- Internal Tables
- Introduction
- Basic Properties of Internal Tables
- Table Keys (Primary, Secondary, Standard, Empty) and Table Indexes
- Creating Internal Tables and Types
- Populating Internal Tables
- Reading Single Lines from Internal Tables
- Determining the Target Area when Reading Single Lines in READ TABLE Statements
- Reading Single Lines by Index
- Reading Single Lines Using Table Keys
- Reading Single Lines Using a Free Key
- Examples of Addressing Individual Components of Read Lines
- Excursions with READ TABLE Statements
- System Field Setting in READ TABLE Statements
- Specifying a WHERE Condition in READ TABLE Statements
- COMPARING and TRANSPORTING Additions: Comparing Fields and Specifying Fields for Transport
- CASTING and ELSE UNASSIGN Additions when Specifying Field Symbols as Target Areas
- BINARY SEARCH Addition: Optimized Read Access When Specifying Free Keys
- Example: Exploring READ TABLE Statements and Table Expressions
- Table Expressions
- Getting Information about Internal Tables, Table Lines, Table Types
- Processing Multiple Internal Table Lines Sequentially
- Operations with Internal Tables Using ABAP SQL SELECT Statements
- Sorting Internal Tables
- Modifying Internal Table Content
- Deleting Internal Table Content
- Grouping Internal Tables
- Collecting Values
- Excursions
- Improving Read Performance with Secondary Table Keys
- Example: Exploring Read Access Performance with Internal Tables
- Generic Table Types with Formal Parameters of Methods and Field Symbols
- Searching and Replacing Substrings in Internal Tables with Character-Like Data Types
- Ranges Tables
- Comparing Content of Compatible Internal Tables
- BDEF Derived Types (ABAP EML)
- More Information
- Executable Example
Internal Tables ...
- are tables that temporarily store variable data (i.e. any number of table lines of a fixed structure) in the working memory in ABAP.
- are dynamic data objects, i.e. all properties apart from the memory consumption are determined statically by the data type.
- consist of a variable sequence of lines of the same data type.
- have a table type as its data type (it is a complex data type), which defines the following properties:
- are used when a variable data set of a random data type needs to be processed in a structured way, for example, storing and processing database table content within an ABAP program.
- allow access to individual table lines via a table index or a table key.
🟢 Click to expand for more details
Line Type
- Defines how each line of the internal table is set up, i. e. it describes what columns the table has.
- It can be any ABAP data type, e.g. an elementary or complex data type as well as a reference type.
- In most cases, the line type is structured. In this case, the individual components of a line are also referred to as the columns of the internal table.
- In a simple case, the line consists of a flat structure with elementary data objects; however, it can also be a deep structure whose components can be structures themselves or even internal tables.
Table Category
- Defines how internal tables are managed and stored internally, and how to access individual table entries.
- Why is it relevant? The use of a suitable table category should meet your requirements, i.e. if the internal tables are large, the different categories can have significant performance differences when accessing table content.
- Note: There are two ways to access internal tables:
- Access by table index: A line of an internal table is accessed by its line number. This kind of access is the fastest way to access table lines.
- Access by table key: A line of an internal table is accessed by searching for specific values in specific columns. Note: The columns in which you search can be key columns, but they can also be non-key columns.
Category | Internally managed by | Access | Primary table key | When to use | Hints |
---|---|---|---|---|---|
STANDARD |
Primary table index (that's why these tables are called index tables) |
|
|
|
|
SORTED |
Primary table index (that's why these tables are called index tables) |
|
... used to sort the table in ascending order. |
|
|
HASHED |
Hash algorithm |
|
Always unique |
|
|
Key Attributes
- There are two types of table keys: a primary table key and secondary table keys.
- Table keys ...
- are intended to provide an optimized access to the content of internal tables.
- are either unique or non-unique, i.e. more than one line with the same key (duplicates) can exist in the internal table or not. Regarding the primary table key, the definition depends on the table category. For the secondary table key, the definition depends on the key type. For standard tables, the primary table key can also be defined as empty, i.e. it does not contain any key columns. Note that for standard tables, an optimized access is only possible with secondary table keys.
- Type of keys:
- Sorted keys:
- Are either the primary table keys of sorted tables or the secondary table keys of any table.
- Are managed internally by a table index. In the case of sorted tables, this is the primary table index. In the case of secondary table keys, a secondary table index is added.
- Access via sorted keys means an optimized binary search.
- Hashed keys:
- Are either the primary table key of hashed tables or secondary table keys of any table.
- Internally managed by a hash algorithm.
- There is no table index for a hashed key.
- Sorted keys:
Further information
🟢 Click to expand for more details
Primary table key
- Each internal table has a primary table key.
- Can be either a self-defined key or the standard key.
- The primary table key is ...
- sorted for sorted tables.
- hashed for hashed tables.
- Note that the key fields in sorted and hashed tables are read-only. This is not valid for standard tables.
- The specification of the primary key can be omitted only for standard tables. The primary table key is then automatically defined as a non-unique standard key.
- The primary table key has the predefined name
primary_key
, by which it can also be addressed explicitly. However, its use is optional, and it is usually not necessary to specify it explicitly. You can also specify an alias name for the primary key. - When accessing internal tables using the table key, the primary key is always used implicitly in processing statements if no secondary key is specified. Note that the primary table key must be specified in table expressions if the primary key is to be used explicitly.
💡 Note
The key can consist of individual key fields or the entire line of the internal table. In this case, the pseudo componenttable_line
can be used to denote the primary table key. For non-structured line types, this is the only way to define the key.
Standard key
- The standard key is a special primary table key.
- It can be declared either explicitly or implicitly.
- Standard key of an internal table with a ...
- structured line type: The primary table key consists of all fields with character-like and byte-like data types.
- non-structured/elementary line type: The entire table is the key (
table_line
).
- An internal table with no explicit key specification implicitly has the standard table key as the primary table key.
- Why respecting standard keys is important:
- Sorting of a table can produce unexpected results.
- Since the standard key can consist of many fields, it affects the performance when accessing the internal table via the keys.
- The key fields of the primary table key of sorted and hashed tables are always read-only, i.e. using the standard key with these table categories and then (unintentionally) modifying fields can cause unexpected runtime errors.
- Specifying keys explicitly has the advantage of making the code more readable and preventing the standard key from being set by mistake.
Empty key
- The primary table key of a standard table can be empty, i.e. it does not contain any key fields.
- An empty key is not possible for sorted and hashed tables.
- When used:
- When the definition of a table key is not important.
- To explicitly state that a table key is not required, instead of specifying no key definition. Otherwise, the standard key is used, which can lead to unexpected results as mentioned above.
- Declaration:
- Explicit declaration using the addition
EMPTY KEY
. - Implicit declaration when using the standard key if a structured line type does not contain non-numeric elementary components or if an unstructured line type is tabular.
- Note: When using an inline declaration such as
... INTO TABLE @DATA(itab) ...
inSELECT
statements, the resulting table is a standard table and has an empty key.
- Explicit declaration using the addition
💡 Primary table index
- Index tables, such as sorted and standard tables, have a primary table index.
- Each table line receives a unique line number from the primary table index.
- The index updates each time a line is added or removed.
Secondary table keys
- Secondary table keys ...
- are optional for all table categories.
- can be unique/non-unique sorted keys or unique hash keys.
- have a self-defined name. An alias name can also be specified.
- A secondary table index is created internally for each sorted secondary key. This allows index access to hashed tables via the secondary table key. In this case,
sy-tabix
is set. - When accessing internal tables using the secondary table key, the key name (or the alias if specified) must be specified. They are not selected automatically. If no secondary key is specified in a processing statement, the primary key or primary table index is always used. If you want to make use of this key in ABAP statements, for example,
READ
,LOOP AT
orMODIFY
statements, you must specify the key explicitly using the appropriate additions, for example,WITH ... KEY ... COMPONENTS
orUSING KEY
. - Use cases:
- To improve read performance.
- For very large internal tables (that are populated once and changed very often)
- Standard tables, whose primary table keys cannot be unique, can be provided with a means of ensuring that unique table entries are read.
- Note that defining secondary table keys involves additional administration costs (additional memory consumption). Therefore, the use of secondary table keys should be reserved for cases where the benefits outweigh the extra costs.
- For more details, see the programming guidelines for secondary keys: Secondary Key (F1 docu for standard ABAP).
💡 Note
- See examples of internal table declarations using the table keys mentioned above in the following section.
- See more information and examples that focus on table access using keys and index below.
You can declare internal tables and internal table types in ABAP programs using the TYPES
and DATA
statements. The relevant syntax elements for internal tables are TABLE OF
in combination
with the additions TYPE
or LIKE
.
TYPES itab_type1 TYPE STANDARD TABLE OF data_type ... "Standard table type
TYPES itab_type2 LIKE SORTED TABLE OF data_object ... "Sorted table type
DATA itab1 TYPE TABLE OF data_type ... "Standard table by default
DATA itab2 TYPE HASHED TABLE OF data_type ... "Hashed table
DATA itab3 TYPE itab_type1 ... "Based on an existing internal table type
DATA itab4 LIKE itab1 ... "Based on an existing internal table
💡 Note
- If the table category is not specified (
... TYPE TABLE OF ...
), it is automatically... TYPE STANDARD TABLE OF ...
.- Using Runtime Type Creation (RTTC), you can define and create new internal tables and table types as type description objects at runtime. For more information, see the Dynamic Programming ABAP cheat sheet.
The following code snippets contain various internal table declarations. It is intended to demonstrate a selection of the rich variety of possible internal tables mentioned in the previous sections, e.g. in Table Keys in Internal Tables. In the examples, many of the internal tables are created using the structured type of a demo database table in the DDIC. The line type of the database table is automatically used when defining an internal table.
Creating internal tables with ... | Details/Code Snippet |
Standard table keys |
"Standard table without explicit primary table key specification.
"Note that STANDARD is not explicitly specified.
"Implicitly, the standard key is used; all non-numeric table fields
"make up the primary table key.
DATA it1 TYPE TABLE OF zdemo_abap_flsch.
"Explicitly specifying STANDARD for a standard table.
"Explicitly specifying the standard table key. The declaration
"corresponds to it1.
DATA it2 TYPE STANDARD TABLE OF zdemo_abap_flsch WITH DEFAULT KEY.
"Hashed table with unique standard table key
DATA it3 TYPE HASHED TABLE OF zdemo_abap_flsch WITH UNIQUE DEFAULT KEY.
"Sorted table with non-unique standard table key
DATA it4 TYPE SORTED TABLE OF zdemo_abap_flsch WITH NON-UNIQUE DEFAULT KEY. |
Primary table keys |
"Specifying the primary table key
"In standard tables, only a non-unique key is possible.
"The following two examples are the same. NON-UNIQUE can be ommitted but
"is implicitly added.
DATA it5 TYPE TABLE OF zdemo_abap_flsch WITH NON-UNIQUE KEY carrid.
DATA it6 TYPE TABLE OF zdemo_abap_flsch WITH KEY carrid.
"Sorted tables: both UNIQUE and NON-UNIQUE possible
DATA it7 TYPE SORTED TABLE OF zdemo_abap_flsch WITH UNIQUE KEY carrid connid.
DATA it8 TYPE SORTED TABLE OF zdemo_abap_flsch WITH NON-UNIQUE KEY carrid connid cityfrom.
"Hashed tables: UNIQUE KEY must be specified
DATA it9 TYPE HASHED TABLE OF zdemo_abap_flsch WITH UNIQUE KEY carrid.
"Explicitly specifying the predefined name primary_key and listing the components.
"The example is the same as it5 and it6.
DATA it10 TYPE TABLE OF zdemo_abap_flsch WITH KEY primary_key COMPONENTS carrid.
"The following example is the same as it8.
DATA it11 TYPE SORTED TABLE OF zdemo_abap_flsch
WITH NON-UNIQUE KEY primary_key COMPONENTS carrid connid cityfrom.
"Specifying an alias name for a primary table key.
"Only possible for sorted/hashed tables.
DATA it12 TYPE SORTED TABLE OF zdemo_abap_flsch
WITH NON-UNIQUE KEY primary_key
ALIAS p1 COMPONENTS carrid connid cityfrom.
"Specifying a key that is composed of the entire line using the
"predefined table_line.
"In the example, an alias name is defined for a primary table key.
DATA it13 TYPE HASHED TABLE OF zdemo_abap_flsch
WITH UNIQUE KEY primary_key
ALIAS p2 COMPONENTS table_line. |
Empty keys |
"Empty keys are only possible for standard tables
DATA it14 TYPE TABLE OF zdemo_abap_flsch WITH EMPTY KEY.
"Empty primary table key, secondary table key specified
DATA it15 TYPE TABLE OF zdemo_abap_flsch WITH EMPTY KEY
WITH UNIQUE SORTED KEY cities COMPONENTS cityfrom cityto.
"Excursion: The inline declaration in a SELECT statement
"produces a standard table with empty key.
SELECT * FROM zdemo_abap_flsch INTO TABLE @DATA(it16). |
Secondary table keys |
"The following examples demonstrate secondary table keys that are
"possible for all table categories.
DATA it17 TYPE TABLE OF zdemo_abap_flsch "standard table
WITH NON-UNIQUE KEY carrid connid "primary table key
WITH UNIQUE SORTED KEY cities COMPONENTS cityfrom cityto. "secondary table key
DATA it18 TYPE HASHED TABLE OF zdemo_abap_flsch "hashed table
WITH UNIQUE KEY carrid connid
WITH NON-UNIQUE SORTED KEY airports COMPONENTS airpfrom airpto.
DATA it19 TYPE SORTED TABLE OF zdemo_abap_flsch "sorted table
WITH UNIQUE KEY carrid connid
WITH UNIQUE HASHED KEY countries COMPONENTS countryfr countryto.
"Multiple secondary keys are possible
DATA it20 TYPE TABLE OF zdemo_abap_flsch
WITH NON-UNIQUE KEY primary_key COMPONENTS carrid connid
WITH NON-UNIQUE SORTED KEY cities COMPONENTS cityfrom cityto
WITH UNIQUE HASHED KEY airports COMPONENTS airpfrom airpto.
"Alias names for secondary table keys (and also for the primary
"table key in the example)
DATA it21 TYPE SORTED TABLE OF zdemo_abap_flsch
WITH NON-UNIQUE KEY primary_key ALIAS k1 COMPONENTS carrid connid
WITH NON-UNIQUE SORTED KEY cities ALIAS s1 COMPONENTS cityfrom cityto
WITH UNIQUE HASHED KEY airports ALIAS s2 COMPONENTS airpfrom airpto.
"Excursion: Example of using table keys and alias names using a LOOP AT statement.
"All of the statements below are possible.
"Note that if the secondary table key is not specified (and if the USING KEY addition is not
"used in the example), the primary table key is respected by default.
"Further ABAP statements, such as READ or MODIFY, are available in which the key can be
"explicitly specified to process internal tables.
LOOP AT it21 INTO DATA(wa) USING KEY primary_key.
"LOOP AT it21 INTO DATA(wa) USING KEY k1.
"LOOP AT it21 INTO DATA(wa) USING KEY cities.
"LOOP AT it21 INTO DATA(wa) USING KEY s1.
"LOOP AT it21 INTO DATA(wa) USING KEY airports.
"LOOP AT it21 INTO DATA(wa) USING KEY s2.
...
ENDLOOP. |
The examples above demonstrate internal tables that are created using the structured type of a database table in the DDIC. The following example shows the pattern and various examples of declaring internal tables and types by including the local definition of structured data and internal table types for demonstration purposes.
Steps:
- Define a structured data type (locally or globally). This is not necessary if you use, for example, the name of a database table or CDS view in the internal table declaration. In these cases their line type is used automatically.
- Define an internal table type.
- Create the internal table, i.e. a data object that uses this type.
You can also create an internal table by ...
- combining the data object creation and table type definition in one step.
- using an inline declaration. Such inline declarations are possible at suitable declaration
positions
if the operand type can be fully determined, for example, using a
DATA
statement (orFINAL
for immutable variables).
"Internal table creation steps based on a locally declared structure
"1. Defining line type locally
TYPES: BEGIN OF ls_loc,
key_field TYPE i,
char1 TYPE c LENGTH 10,
char2 TYPE c LENGTH 10,
num1 TYPE i,
num2 TYPE i,
END OF ls_loc.
"2. Defining internal table types
TYPES: tt_std TYPE TABLE OF ls_loc WITH EMPTY KEY,
tt_sorted TYPE SORTED TABLE OF ls_loc WITH NON-UNIQUE KEY key_field,
tt_hashed TYPE HASHED TABLE OF ls_loc WITH UNIQUE KEY key_field.
"3. Creating internal tables from locally defined table types
DATA: it22 TYPE tt_std,
it23 TYPE tt_sorted,
it24 TYPE tt_hashed.
"Combining data object and table type definition
DATA it25 TYPE TABLE OF ls_loc WITH NON-UNIQUE KEY key_field.
"LIKE addition
"Internal table based on an already existing internal table using LIKE.
DATA it26 LIKE it25.
"In the following example, an internal table is created containing internal tables
"of the type of itab_a6.
DATA it27 LIKE TABLE OF it25.
This section explores various line and table type options when declaring internal tables. Examples have already been covered in the previous sections. Among the options are, for example:
- Elementary line types based on elementary built-in ABAP types, locally declared elementary types, globally available elementary types such as DDIC data elements
- Line types based on both locally declared and globally available structured types
- Among the globally available line types are, for example, DDIC structures, database tables and CDS objects such as CDS view entities
- Note that internal tables can be created as deep tables. That is, internal tables can contain deep components (references, internal tables or strings). Structures can also be components of internal tables. For more information, see the Variants of Structures in the Structures cheat sheet.
- Table types based on both locally declared and globally available table types
- References
💡 Note
Types declared in the public visibility section of classes/interfaces are also globally visible and can be used for the creation.
Subject | Details/Code Snippet |
Elementary line types |
"------ Internal tables declared using elementary line types ------
"Note: In this case, the whole table line is the standard table key.
"Elementary built-in ABAP types
DATA it_elem_1 TYPE TABLE OF i.
DATA it_elem_2 TYPE TABLE OF string WITH EMPTY KEY.
DATA it_elem_3 TYPE TABLE OF xstring WITH EMPTY KEY.
DATA it_elem_4 TYPE TABLE OF utclong WITH EMPTY KEY.
"Note: As described above, the table key can consist of components of the
"line type or of the entire line using the pseudo component table_line. In
"case of elementary line types, table_line is the only option to refer to the
"component/entire line.
DATA it_elem_5 TYPE SORTED TABLE OF string WITH NON-UNIQUE KEY table_line.
"Locally declared elementary types
TYPES ty_c3 TYPE c LENGTH 3.
DATA it_elem_6 TYPE TABLE OF ty_c3 WITH EMPTY KEY.
"Apart from the globally available elementary built-in ABAP types, you can
"also use elementary types from the DDIC such as data elements that are
"visible anywhere. You can also refer to CDS simple types that
"constitute elemenary data types.
DATA it_elem_7 TYPE TABLE OF timestamp WITH EMPTY KEY.
DATA it_elem_8 TYPE TABLE OF timestampl WITH EMPTY KEY.
"Globally available elementary types in the public visibility section of
"classes/interfaces
"Elementary type declared in an interface
DATA it_elem_9 TYPE TABLE OF zdemo_abap_get_data_itf=>occ_rate WITH EMPTY KEY.
"Note the syntax options when referring to components. In the following
"example, a database table field is referred to using the database table name,
"the component selector and the name of a field having an elementary data type.
DATA it_elem_10 TYPE TABLE OF zdemo_abap_fli-carrid WITH EMPTY KEY.
"CDS view entity component
DATA it_elem_11 TYPE TABLE OF zdemo_abap_carr_ve-url WITH EMPTY KEY. |
Structured types |
"------ Internal tables declared using structured types ------
"Locally declared structured type
TYPES: BEGIN OF local_struct,
comp1 TYPE i,
comp2 TYPE string,
comp3 TYPE c LENGTH 3,
END OF local_struct.
DATA it_struc_1 TYPE SORTED TABLE OF local_struct WITH UNIQUE KEY comp1.
"Globally available structured types
"E.g. DDIC database tables
DATA it_struc_2 TYPE TABLE OF zdemo_abap_fli WITH EMPTY KEY.
"CDS objects such as ...
"CDS view entity
DATA it_struc_3 TYPE TABLE OF zdemo_abap_carr_ve WITH EMPTY KEY.
"CDS abstract entity
DATA it_struc_4 TYPE TABLE OF zdemo_abap_abstract_ent WITH EMPTY KEY.
"CDS table function
DATA it_struc_5 TYPE TABLE OF zdemo_abap_table_function WITH EMPTY KEY.
"Globally available structured types in the public visibility section of
"classes/interfaces
DATA it_struc_6 TYPE TABLE OF zcl_demo_abap_amdp=>fli_struc WITH EMPTY KEY.
"Many of the previous examples demonstrate flat line types.
"Deep line types are also possible, that is, the line type
"can include components such as strings, references, and internal
"tables.
TYPES: BEGIN OF loc_deep_struct,
key_field TYPE i,
char TYPE c LENGTH 10,
str TYPE string,
dref TYPE REF TO i,
struct TYPE zdemo_abap_flsch,
str_tab TYPE string_table,
tab TYPE TABLE OF zdemo_abap_flsch WITH EMPTY KEY,
END OF loc_deep_struct.
DATA it_struc_7 TYPE TABLE OF loc_deep_struct WITH EMPTY KEY. |
Table types |
"------ Internal tables declared using table types ------
"Locally declared table type (based on a locally declared structured type)
TYPES: BEGIN OF loc_struct,
comp1 TYPE i,
comp2 TYPE string,
comp3 TYPE c LENGTH 3,
END OF loc_struct,
ty_tab_1 TYPE TABLE OF loc_struct WITH EMPTY KEY.
DATA it_tab_1 TYPE ty_tab_1.
"Locally declared table type (based on a globally available structured type)
TYPES ty_tab_2 TYPE TABLE OF zdemo_abap_fli WITH EMPTY KEY.
DATA it_tab_2 TYPE ty_tab_2.
"Globally available table types
"The following examples show released table types. They have an
"elementary line type.
DATA it_tab_3 TYPE string_table.
DATA it_tab_4 TYPE string_hashed_table.
DATA it_tab_5 TYPE xstring_table.
"Globally available table types in the public visibility section of
"classes/interfaces
DATA it_tab_6 TYPE zdemo_abap_get_data_itf=>carr_tab.
DATA it_tab_7 TYPE zcl_demo_abap_amdp=>flsch_tab.
"Creating an internal table based on the table type of an existing
"internal table
DATA it_tab_8 LIKE it_tab_7. |
References as line types |
DATA it_ref_1 TYPE TABLE OF REF TO i.
DATA it_ref_2 TYPE TABLE OF REF TO string.
"Generic type data
DATA it_ref_3 TYPE TABLE OF REF TO data.
"Such a table can hold any data type
"For more details on generic types, see the Data Types and Data Objects
"and Dynamic Programming cheat sheets.
it_ref_3 = VALUE #( ( NEW i( 3 ) ) "Elementary type
( NEW string( `hello` ) ) "Elementary type
( NEW zdemo_abap_flsch( carrid = 'XY' connid = '1234' ) ) "Structured type
( NEW string_table( ( `a` ) ( `b` ) ( `c` ) ) ) "Table type
). |
"Table declared inline in the context of an assignment
"The examples show the copying of a table including the content
"on the fly and creating the table in one step. The data type of the
"declared variable is determined by the right side.
DATA string_tab TYPE string_table.
DATA(it_a) = string_tab.
DATA(it_b) = it_a.
"Using FINAL for creating immutable variables
FINAL(it_c) = it_a.
"For example, it is not possible to modify such a table in the following position.
"Find more information in the Data Types and Data Objects cheat sheet.
"APPEND INITIAL LINE TO it_c.
"As shown below and in other cheat sheets, constructor operators
"are handy when creating internal tables in place. The following
"examples uses the VALUE operator and an internal table type.
DATA(it_d) = VALUE string_table( ( `aaa` )
( `bbb` ) ).
"Not providing any table lines means the table is initial
"and has the same effect as the declaration of it_f.
DATA(it_e) = VALUE string_table( ).
DATA it_f TYPE string_table.
"Excursion
"Table declared inline in the context of a SELECT statement;
"a prior extra declaration of an internal table is not needed.
SELECT * FROM zdemo_abap_fli INTO TABLE @DATA(it_g).
"Instead of
DATA it_h TYPE TABLE OF zdemo_abap_fli WITH EMPTY KEY.
SELECT * FROM zdemo_abap_fli INTO TABLE @it_h.
"Using FINAL
SELECT * FROM zdemo_abap_fli INTO TABLE @FINAL(it_i).
Find more information in the Dynamic Programming cheat sheet.
DATA(some_type) = 'STRING'.
DATA dataref TYPE REF TO data.
"Creating an internal table using a CREATE DATA statement
"by specifying the type name dynamically.
"In the example, a standard table with elementary line type
"and standard key is created.
CREATE DATA dataref TYPE TABLE OF (some_type).
TYPES: BEGIN OF demo_struc,
comp1 TYPE c LENGTH 10,
comp2 TYPE i,
comp3 TYPE i,
END OF demo_struc.
"Internal table with structured line type and empty key.
CREATE DATA dataref TYPE TABLE OF ('DEMO_STRUC') WITH EMPTY KEY.
"Using a globally available table type
CREATE DATA dataref TYPE ('STRING_TABLE').
To copy internal table content from one table to another, you can use the assignment operator. Such an assignment (without a constructor expression) deletes the existing content in the target internal table. The example below assumes that the source and target table have compatible line types. Using inline declaration is helpful to avoid an additional internal table declaration with an appropriate type.
itab = itab2.
DATA(itab3) = itab.
FINAL(itab4) = itab.
💡 Note
- Internal tables can only be assigned to internal tables.
- Internal tables can be assigned to each other if their line types are compatible or convertible.
- An assignment can trigger an uncatchable exception if, for example, the target table is assigned a duplicate of a unique primary table key or secondary table.
- More information: Conversion Rules for Internal Tables
You can use the ABAP keywords
INSERT
and APPEND
to add lines to internal tables.
🟢 Click to expand for notes on using APPEND
and INSERT
APPEND
...- always adds lines at the bottom of the internal table.
- is not a problem for standard tables where lines are managed
by an index. When the statement is used, the system field
sy-tabix
is set to the index of the recently added line.sy-tabix
is always set to the index with respect to the primary table index. - cannot be used for hashed tables. For sorted tables,
lines are appended only if they match the sort order, and no
duplicate entries are created if the primary table key is unique.
Therefore,
INSERT
should be used when adding lines to sorted tables.
INSERT
...- can be used to add lines at a specific position in tables (by specifying the target index). In doing so, all the following lines are moved down one position.
- without specifying the position adds the lines at the bottom of
the table in case of standard tables. However, when using
INSERT
,sy-tabix
is not set unlikeAPPEND
. In case of sorted tables, the line is automatically inserted at the correct position. - With
INSERT
, you can specify the position in the internal table at which lines are inserted afterINTO
.... INTO TABLE itab ...
: Line is inserted in ...- standard tables as last line (i.e. appended)
- sorted tables in the sort order in accordance with primary key values
- hashed table by the hash administration in accordance with primary key values
... INTO itab INDEX n
: Possible for index tables. The line is inserted before the line with the line numbern
in the primary table index.
- Note: In the case of unique primary table keys, the table cannot have entries with duplicate
keys. If a duplicate is inserted, the insertion fails and the
system field
sy-subrc
is set to 4.
- What to use? The recommendation is the
INSERT
statement. It covers all table and key types. Consider potential issues when you change table/key types, and you useAPPEND
in your code.
Subject | Details/Code Snippet |
Adding a line to an internal table |
The example shows both a structure that is created using the APPEND VALUE #( comp1 = a comp2 = b ... ) TO itab.
APPEND struc TO itab.
INSERT VALUE #( comp1 = a comp2 = b ... ) INTO TABLE itab.
INSERT struc INTO TABLE itab. |
Adding an initial line |
... to an internal table without providing any field values.
APPEND INITIAL LINE TO itab.
INSERT INITIAL LINE INTO TABLE itab. |
Adding a line and assigning the added line to a field symbol or data reference variable |
"When inserting single lines, you can specify the optional additions
"ASSIGNING and REFERENCE INTO. If the insertion is successful, the
"line is assigned to a field symbol or a data reference variable.
"The targets can also be created inline.
APPEND VALUE #( comp1 = a comp2 = b ... ) TO itab ASSIGNING FIELD-SYMBOL(<fs>).
APPEND INITIAL LINE TO itab ASSIGNING <fs>.
INSERT INITIAL LINE INTO TABLE itab REFERENCE INTO DATA(dref).
"Note: Once assigned, you can then, for example, address individual components
"of the (initial) line and assign values, e.g. <fs>-comp = ... or dref->comp = .... |
Adding all lines from another internal table |
APPEND LINES OF itab2 TO itab.
INSERT LINES OF itab2 INTO TABLE itab. |
Adding multiple lines from another internal table with a specified index range |
APPEND LINES OF itab2 FROM 3 TO 5 TO itab.
APPEND LINES OF itab2 FROM 3 TO itab.
APPEND LINES OF itab2 TO 7 TO itab.
INSERT LINES OF itab2 FROM 5 TO 9 INTO TABLE itab. |
Inserting one line or multiple lines from another internal table at a specific position |
INSERT struc INTO itab2 INDEX i.
INSERT LINES OF itab2 INTO itab INDEX i. |
The constructor expressions can be specified in/with various positions/statements in ABAP. In most of the following snippets, simple assignments are demonstrated.
💡 Note
The following sections cover a selection. There are more constructor expressions used in the context of internal tables (e.g. for creating internal tables). Find more details in the Constructor Expressions ABAP cheat sheet.
As mentioned above, table lines that are constructed inline as
arguments to the VALUE
operator, for example, can be added to
internal tables. In the following cases, internal tables are populated
using constructor expressions in the context of
assignments.
Subject | Details/Code Snippet |
Populating an existing internal table by assigning an internal table that is constructed inline |
In the example below, the internal table is populated by assigning an
internal table that is constructed inline with the
itab = VALUE #( ( line )
( comp1 = a comp2 = b ... ) ). |
Creating an internal table by inline declaration and adding lines with a constructor expression |
"Internal table type
TYPES it_type LIKE itab.
"Inline declaration
"The # character would not be possible here since the line type
"cannot be derived from the context.
DATA(it_in) = VALUE it_type( ( comp1 = a comp2 = b ... )
( comp1 = c comp2 = d ... ) ).
"Creating string tables
DATA(str_tab_a) = VALUE string_table( ( `Hallo` ) ( `World` ) ).
DATA(str_tab_b) = VALUE string_table( ).
"In the previous statement, the internal table is declared
"inline, however, no content, no table lines are provided.
"This means that an initial string table was created. This
"way, the statement has the same effect as the following
"statement.
DATA str_tab_c TYPE string_table. |
Adding new lines without deleting existing content |
When you use the above assignments to an existing internal table ( itab = VALUE #( BASE itab ( comp1 = a comp2 = b ... )
( comp1 = c comp2 = d ... ) ). |
Adding lines of other tables |
... using the itab = VALUE #( ( comp1 = a comp2 = b ...)
( comp1 = a comp2 = b ...)
"All lines
( LINES OF itab2 )
"More syntax options
( LINES OF itab3 FROM 2 TO 5 )
( LINES OF itab4 FROM 3 )
( LINES OF itab5 TO 7 )
( LINES OF itab6 STEP 2 )
( LINES OF itab7 USING KEY primary_key FROM 3 TO 6 )
... ). |
Subject | Details/Code Snippet |
Copying the content of another internal table |
... using the
itab = CORRESPONDING #( itab3 ).
MOVE-CORRESPONDING itab3 TO itab. |
Copying content and retaining existing content |
... using the itab = CORRESPONDING #( BASE ( itab ) itab3 ).
MOVE-CORRESPONDING itab3 TO itab KEEPING TARGET LINES. |
Assigning components using mapping relationships |
itab = CORRESPONDING #( itab3 MAPPING a = c b = d ). |
Excluding components from the assignment |
... using the
itab = CORRESPONDING #( itab3 EXCEPT e ).
itab = CORRESPONDING #( itab3 EXCEPT * ). |
Preventing runtime errors when duplicate lines are assigned |
... to the target table that is defined to accept only unique keys using the
itab = CORRESPONDING #( itab2 DISCARDING DUPLICATES ). |
Copying data from deep internal tables |
itab_nested2 = CORRESPONDING #( DEEP itab_nested1 ).
itab_nested2 = CORRESPONDING #( DEEP BASE ( itab_nested2 ) itab_nested1 ).
MOVE-CORRESPONDING itab_nested1 TO itab_nested2 EXPANDING NESTED TABLES.
MOVE-CORRESPONDING itab_nested1 TO itab_nested2 EXPANDING NESTED TABLES KEEPING TARGET LINES. |
💡 Note
TheCL_ABAP_CORRESPONDING
class can be used for assignments. Find an example in the Misc ABAP Classes cheat sheet.
To create an internal table by copying data from another internal table and
filtering out lines that do not meet the WHERE
condition, you can use the FILTER
operator.
- The
FILTER
operator constructs an internal table according to a specified type (which can be an explicitly specified, non-generic table type or the#
character as a symbol for the operand type before the first parenthesis). - The lines for the new internal table are taken from an
existing internal table based on conditions specified in a
WHERE
clause. Note that the table type of the existing internal table must be convertible into the specified target type. - The conditions can be based on either single values or a filter table.
- Additions:
Addition | Details |
---|---|
USING KEY |
Specifies the table key used to evaluate the WHERE condition: either a sorted key or a hash key. If the internal table does not have either of these, it must have a secondary table key, which must be specified after USING KEY . |
EXCEPT |
Specifying EXCEPT means that those lines of the existing table are used that do not meet the condition specified in the WHERE clause. If EXCEPT is not specified, those lines of the existing table that meet the condition are used. |
Examples:
"FILTER on conditions based on single values
"Assumption: The component num is of type i.
DATA itab1 TYPE SORTED TABLE OF struc WITH NON-UNIQUE KEY num.
DATA itab2 TYPE STANDARD TABLE OF struc WITH NON-UNIQUE SORTED KEY sec_key COMPONENTS num.
DATA itab3 TYPE HASHED TABLE OF struc WITH UNIQUE KEY num.
"The lines meeting the condition are respected.
"Note: The source table must have at least one sorted or hashed key.
"Here, the primary key is used
DATA(f1) = FILTER #( itab1 WHERE num >= 3 ).
"USING KEY primary_key explicitly specified; same as above
DATA(f2) = FILTER #( itab1 USING KEY primary_key WHERE num >= 3 ).
"EXCEPT addition
DATA(f3) = FILTER #( itab1 EXCEPT WHERE num >= 3 ).
DATA(f4) = FILTER #( itab1 EXCEPT USING KEY primary_key WHERE num >= 3 ).
"Secondary table key specified after USING KEY
DATA(f5) = FILTER #( itab2 USING KEY sec_key WHERE num >= 4 ).
DATA(f6) = FILTER #( itab2 EXCEPT USING KEY sec_key WHERE num >= 3 ).
"Note: In case of a hash key, exactly one comparison expression for each key component is allowed;
"only = as comparison operator possible.
DATA(f7) = FILTER #( itab3 WHERE num = 3 ).
"Using a filter table
"In the WHERE condition, the columns of source and filter table are compared.
"Those lines in the source table are used for which at least one line in the filter
"table meets the condition. EXCEPT and USING KEY are also possible.
DATA filter_tab1 TYPE SORTED TABLE OF i
WITH NON-UNIQUE KEY table_line.
DATA filter_tab2 TYPE STANDARD TABLE OF i
WITH EMPTY KEY
WITH UNIQUE SORTED KEY line COMPONENTS table_line.
DATA(f8) = FILTER #( itab1 IN filter_tab1 WHERE num = table_line ).
"EXCEPT addition
DATA(f9) = FILTER #( itab1 EXCEPT IN filter_tab1 WHERE num = table_line ).
"USING KEY is specified for the filter table
DATA(f10) = FILTER #( itab2 IN filter_tab2 USING KEY line WHERE num = table_line ).
"USING KEY is specified for the source table, including EXCEPT
DATA(f11) = FILTER #( itab2 USING KEY sec_key EXCEPT IN filter_tab2 WHERE num = table_line ).
💡 Note
More constructor expressions are available to deal with internal tables, for example theREDUCE
operator. Find more information and examples in the Constructor Expression cheat sheet.
Using the instance operator NEW
, you can create anonymous data objects, such as anonymous internal tables. You can access the lines, components or the entire data objects by dereferencing. For more information, refer to the Dynamic Programming and Constructor Expressions cheat sheets.
TYPES: BEGIN OF s,
a TYPE c LENGTH 3,
b TYPE i,
END OF s,
tab_type TYPE TABLE OF s WITH EMPTY KEY.
"Creating and populating an anonymous data object
DATA(dref_tab) = NEW tab_type( ( a = 'aaa' b = 1 )
( a = 'bbb' b = 2 ) ).
"Access by derefencing
DATA(copy_deref_itab) = dref_tab->*.
DATA(read_line) = dref_tab->*[ 2 ].
DATA(read_comp) = dref_tab->*[ 1 ]-a.
dref_tab->*[ 1 ]-a = 'zzz'.
ASSERT dref_tab->*[ 1 ]-a = 'zzz'.
INSERT VALUE s( a = 'yyy' b = 3 ) INTO TABLE dref_tab->*.
Expand the following collapsible section for example code. To try it out, create a demo class named zcl_some_class
and paste the code into it. After activation, choose F9 in ADT to execute the class.
The example is set up to display output in the console, but only for few data objects. You may want to set a break point at the earliest possible position and walk through the example in the debugger. This will allow you to double-click on data objects and observe how the different statements affect their contents.
🟢 Click to expand for a code example
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
"Populating internal tables by adding a line (structure) using
"APPEND ... TO/INSERT ... INTO
TYPES: BEGIN OF st_a,
num TYPE i,
str TYPE string,
char TYPE c LENGTH 2,
END OF st_a,
ty_tab_a TYPE TABLE OF st_a WITH EMPTY KEY.
DATA it_a TYPE ty_tab_a.
"Adding a line created inline
APPEND VALUE #( num = 1 str = `A` char = 'bb' ) TO it_a.
INSERT VALUE #( num = 2 str = `C` char = 'dd' ) INTO TABLE it_a.
"Adding an existing line
DATA(struc_a) = VALUE st_a( num = 3 str = `E` char = 'ff' ).
"Structure whose components are assigned individually using the
"structure component selector
DATA struc_b TYPE st_a.
struc_b-num = 4.
struc_b-str = `G`.
struc_b-char = 'hh'.
APPEND struc_a TO it_a.
INSERT struc_b INTO TABLE it_a.
**********************************************************************
"INITIAL LINE addition
"Adding an initial line
APPEND INITIAL LINE TO it_a.
INSERT INITIAL LINE INTO TABLE it_a.
**********************************************************************
"ASSIGNING/REFERENCE INTO additions
"Adding a line and assigning the added line to a field symbol or
"data reference variable
"Creating field symbol inline
APPEND VALUE st_a( num = 5 str = `I` char = 'jj' ) TO it_a ASSIGNING FIELD-SYMBOL(<fs_a>).
"Addressing individual components
ASSERT <fs_a>-num = 5.
<fs_a>-num = 123.
FIELD-SYMBOLS <fs_b> TYPE st_a.
DATA(struc_c) = VALUE st_a( num = 6 ).
INSERT struc_c INTO TABLE it_a ASSIGNING <fs_b>.
<fs_b>-str = `K`.
"Adding an initial line
"The examples use data reference variables.
"Using inline declaration
APPEND INITIAL LINE TO it_a REFERENCE INTO DATA(dref_a).
dref_a->num = 7.
DATA dref_b TYPE REF TO st_a.
INSERT INITIAL LINE INTO TABLE it_a REFERENCE INTO dref_b.
dref_b->num = 8.
DO 3 TIMES.
APPEND INITIAL LINE TO it_a REFERENCE INTO dref_b.
dref_b->* = VALUE #( num = sy-index str = sy-index char = sy-index ).
ENDDO.
**********************************************************************
"LINES OF addition
"Adding all lines from another internal table
"Adding lines to one internal table that are all added to
"another one
DATA it_b TYPE ty_tab_a.
INSERT VALUE #( num = 99 str = `L` char = 'mm' ) INTO TABLE it_b.
INSERT VALUE #( num = 100 str = `N` char = 'oo' ) INTO TABLE it_b.
APPEND LINES OF it_b TO it_a.
INSERT LINES OF it_b INTO TABLE it_a.
**********************************************************************
"Adding multiple lines from another internal table with
"a specified index range
APPEND LINES OF it_a FROM 5 TO 7 TO it_b.
APPEND LINES OF it_a FROM 12 TO it_b. "further lines up to the last line
APPEND LINES OF it_a TO 3 TO it_b. "all lines from the start up to the specified index
INSERT LINES OF it_a FROM 8 TO 10 INTO TABLE it_b.
**********************************************************************
"INDEX addition
"Inserting one line or multiple lines from another internal table at
"a specific position. To be used for index tables.
INSERT VALUE #( num = 9 str = `P` char = 'qq' ) INTO it_b INDEX 2.
INSERT LINES OF VALUE ty_tab_a( ( num = 10 str = `R` ) ( num = 11 str = `S` ) ) INTO it_b INDEX 5.
"FROM and TO can also be used
INSERT LINES OF it_a FROM 1 TO 3 INTO it_b INDEX 1.
**********************************************************************
"Using the VALUE operator
DATA(struc_d) = VALUE st_a( num = 11 str = `T` char = 'uu' ).
DATA it_c TYPE ty_tab_a.
"Adding an existing line and a line created inline
it_c = VALUE #( ( struc_d )
( num = 11 str = `V` char = 'ww' ) ).
"Creating an internal table by inline declaration and adding lines with VALUE
DATA(it_d) = VALUE ty_tab_a( ( num = 12 str = `X` char = 'yy' )
( num = 13 str = `Z` char = 'aa' )
( struc_d ) ).
"******* BASE addition *******
"Adding new lines without deleting existing content
it_d = VALUE #( BASE it_d ( num = 14 str = `B` char = 'cc' )
( num = 15 str = `D` char = 'ee' ) ).
"******* LINES OF addition *******
"Adding lines of other tables
it_d = VALUE #( ( LINES OF it_c ) ). "No BASE addition, existing content is deleted
it_c = VALUE #( BASE it_c ( num = 16 str = `F` char = 'gg' )
( LINES OF it_d ) ).
**********************************************************************
"CORRESPONDING operator / MOVE-CORRESPONDING statements
"Creating and populating demo internal tables
TYPES: BEGIN OF st_b,
num TYPE i,
char TYPE c LENGTH 2,
comp_a TYPE string,
END OF st_b,
ty_tab_b TYPE TABLE OF st_b WITH EMPTY KEY,
BEGIN OF st_c,
num TYPE i,
char TYPE c LENGTH 2,
comp_b TYPE string,
END OF st_c,
ty_tab_c TYPE TABLE OF st_c WITH EMPTY KEY.
DATA(it_e_original) = VALUE ty_tab_b( ( num = 1 char = 'aa' comp_a = `B` )
( num = 2 char = 'cc' comp_a = `D` ) ).
DATA(it_f_original) = VALUE ty_tab_c( ( num = 3 char = 'ee' comp_b = `F` )
( num = 4 char = 'gg' comp_b = `H` ) ).
DATA(it_e) = it_e_original.
DATA(it_f) = it_f_original.
"Copying the content of another internal table respecting identically
"named components
"it_f -> it_e
it_e = CORRESPONDING #( it_f ).
"it_e -> it_f
it_e = it_e_original.
MOVE-CORRESPONDING it_e TO it_f.
"******* BASE addition / /KEEPING TARGET LINES addition *******
"Copying content and retaining existing content
it_e = it_e_original.
it_f = it_f_original.
it_e = CORRESPONDING #( BASE ( it_e ) it_f ).
it_e = it_e_original.
it_f = it_f_original.
MOVE-CORRESPONDING it_e TO it_f KEEPING TARGET LINES.
"******* MAPPING addition *******
"Assigning components using mapping relationships
it_e = it_e_original.
it_f = it_f_original.
it_e = CORRESPONDING #( it_f MAPPING comp_a = comp_b ).
it_e = it_e_original.
it_f = CORRESPONDING #( BASE ( it_f ) it_e MAPPING comp_b = comp_a ). "Retaining content with BASE
"******* EXCEPT addition *******
"Excluding components from the assignment
it_e = it_e_original.
it_f = it_f_original.
it_e = CORRESPONDING #( it_f EXCEPT char ).
it_e = it_e_original.
it_f = CORRESPONDING #( it_e MAPPING comp_b = comp_a EXCEPT * ). "Mapping components
"******* DISCARDING DUPLICATES addition *******
"Preventing runtime errors when duplicate lines are assigned
it_e = VALUE #( ( num = 1 char = 'aa' comp_a = `B` )
( num = 1 char = 'cc' comp_a = `D` ) ).
DATA it_g TYPE SORTED TABLE OF st_b WITH UNIQUE KEY num.
"The statement commented out raises the runtime error ITAB_DUPLICATE_KEY.
"it_g = CORRESPONDING #( it_e ).
it_g = CORRESPONDING #( it_e DISCARDING DUPLICATES ).
"******* DEEP addition / EXPANDING NESTED TABLES addition *******
"Handling deep components such as nested internal tables
"Creating and populating demo internal tables
TYPES: BEGIN OF st_d,
char_a TYPE c LENGTH 2,
char_b TYPE c LENGTH 2,
END OF st_d,
BEGIN OF st_e,
char_b TYPE c LENGTH 2,
char_c TYPE c LENGTH 2,
END OF st_e,
BEGIN OF st_f,
comp1 TYPE c LENGTH 2,
comp2 TYPE c LENGTH 2,
comp3 TYPE TABLE OF st_d WITH EMPTY KEY,
END OF st_f,
BEGIN OF st_g,
comp2 TYPE c LENGTH 2,
comp3 TYPE TABLE OF st_e WITH EMPTY KEY,
comp4 TYPE c LENGTH 2,
END OF st_g,
ty_tab_d TYPE TABLE OF st_f WITH EMPTY KEY,
ty_tab_e TYPE TABLE OF st_g WITH EMPTY KEY.
DATA(it_h_original) = VALUE ty_tab_d(
( comp1 = 'a1' comp2 = 'a2' comp3 = VALUE #( ( char_a = 'a3' char_b = 'a4' ) ( char_a = 'a5' char_b = 'a6' ) ) )
( comp1 = 'b1' comp2 = 'b2' comp3 = VALUE #( ( char_a = 'b3' char_b = 'b4' ) ( char_a = 'b5' char_b = 'b6' ) ) ) ).
DATA(it_i_original) = VALUE ty_tab_e(
( comp2 = 'c1' comp3 = VALUE #( ( char_b = 'c2' char_c = 'c3' ) ( char_b = 'c4' char_c = 'c5' ) ) comp4 = 'c6' )
( comp2 = 'd1' comp3 = VALUE #( ( char_b = 'd2' char_c = 'd3' ) ( char_b = 'd4' char_c = 'd5' ) ) comp4 = 'd6' ) ).
DATA(it_h) = it_h_original.
DATA(it_i) = it_i_original.
"Compare the output of the examples
out->write( `******* CORRESPONDING *******` ).
"Note: The following example uses just CORRESPONDING. The outcome of the assignment
"is a different one compared to using DEEP. Refer to the ABAP Keyword Documentation
"for more details.
it_h = CORRESPONDING #( it_i ).
out->write( it_h ).
out->write( `******* CORRESPONDING ... DEEP *******` ).
it_h = CORRESPONDING #( DEEP it_i ).
out->write( it_h ).
out->write( `******* CORRESPONDING ... DEEP BASE *******` ).
it_h = it_h_original.
it_h = CORRESPONDING #( DEEP BASE ( it_h ) it_i ).
out->write( it_h ).
out->write( `******* MOVE-CORRESPONDING ... EXPANDING NESTED TABLES *******` ).
it_h = it_h_original.
MOVE-CORRESPONDING it_i TO it_h EXPANDING NESTED TABLES.
out->write( it_h ).
out->write( `******* MOVE-CORRESPONDING ... EXPANDING NESTED TABLES KEEPING TARGET LINES *******` ).
it_h = it_h_original.
MOVE-CORRESPONDING it_i TO it_h EXPANDING NESTED TABLES KEEPING TARGET LINES.
out->write( it_h ).
ENDMETHOD.
ENDCLASS.
There are three different ways to specify the line to read:
- by index (only index tables)
- by table keys (only tables for which a key is defined)
- by free key
The following code snippets include READ TABLE
statements and table expressions to read from internal tables.
-
Copying a line to a data object using the addition
INTO
. After the copying, the line found exists separately in the internal table and in the data object. If you change the data object or the table line, the change does not affect the other. However, you can modify the copied table line and use aMODIFY
statement to modify the table based on the modified table line (see below). TheTRANSPORTING
addition specifies which components to copy. If it is not specified, all components are respected.READ TABLE itab INTO dobj ... "dobj must have the table's structure type READ TABLE itab INTO DATA(dobj_inl) ... READ TABLE itab INTO ... TRANSPORTING comp1 [comp2 ... ].
-
Assigning a line to a field symbol, for example, using an inline declaration (
... ASSIGNING FIELD-SYMBOL(<fs>) ...
). When you then access the field symbol, it means that you access the found table line. There is no actual copying of content. Therefore, modifying the field symbol means modifying the table line directly. Note that you cannot use theTRANSPORTING
addition since the entire table is assigned to the field symbol.READ TABLE itab ASSIGNING <fs1> ... "The field symbol must have an appropriate type. READ TABLE itab ASSIGNING FIELD-SYMBOL(<fs2>) ... "The field symbol is created inline.
-
Reading a line into a data reference variable using
REFERENCE INTO
. In this case, no copying takes place. If you want to address the line, you must first dereference the data reference. You cannot use the additionTRANSPORTING
.READ TABLE itab REFERENCE INTO dref ... READ TABLE itab REFERENCE INTO DATA(dref_inl) ...
✔️ Hint
Which to use then? Since all syntax options basically provide similar functionality, your use case, the performance or readability of the code may play a role. For more information, see the programming guidelines for the target area (F1 docu for standard ABAP). A use case forINTO dobj
is when the table should not be changed using the copied table line. However, copying comes at a performance cost. Imagine that your table contains many columns or nested components. In this case, it is better not to copy at all (although you can use theTRANSPORTING
addition to restrict the fields to be copied).
Statement | Details/Code Snippet |
READ TABLE |
The following example shows "Primary table index is used by default
READ TABLE itab INTO wa INDEX i.
"Primary table index is used, primary key's default name specified explicitly
READ TABLE itab INTO wa INDEX i USING KEY primary_key.
"Secondary table index is used (assuming a secondary table key sec_key exists)
READ TABLE itab INTO wa INDEX i USING KEY sec_key.
"Note: You can also use alias names for the keys if specified. |
Table expressions |
Using table
expressions,
the read result is stored in a variable that can be declared inline.
The number in the square brackets represents the index. A line that is
not found results in an runtime error. To avoid an error, you can
use a Find more information about syntax related to table expressions further down.
"In the examples, integer values are specified for the index.
DATA(lv1) = itab[ 1 ].
TRY.
DATA(lv2) = itab[ 2 ].
CATCH cx_sy_itab_line_not_found.
...
ENDTRY.
DATA(lv3) = itab[ KEY primary_key INDEX 3 ].
"Copying a table line via table expression and embedding in constructor expression
DATA(lv4) = VALUE #( itab[ 4 ] ).
"Reading into data reference variable using the REF operator
DATA(lv5_ref) = REF #( itab[ 5 ] ). |
Avoiding an exception when using table expressions |
When you read a non-existent line using a table expression, you may not want to throw an exception. You can also embed the table expression
in a constructor expression using the DATA(line1) = VALUE #( itab[ 6 ] OPTIONAL ).
DATA(line2) = VALUE #( itab[ 7 ] DEFAULT itab[ 1 ] ).
DATA(line3) = VALUE #( itab[ 8 ] DEFAULT VALUE #( ) ). |
Lines can be read by explicitly specifying the table keys or the alias names, if any.
Statement | Details/Code Snippet |
READ TABLE |
"Example internal table with primary and secondary table key and alias names
"Assumption: All components are of type i
DATA it TYPE SORTED TABLE OF struc
WITH NON-UNIQUE KEY primary_key ALIAS pk COMPONENTS a b
WITH NON-UNIQUE SORTED KEY sec_key ALIAS sk COMPONENTS c d.
"Primary table key
READ TABLE it INTO wa WITH TABLE KEY primary_key COMPONENTS a = 1 b = 2.
"Alias
READ TABLE it INTO wa WITH TABLE KEY pk COMPONENTS a = 1 b = 2.
"Secondary table key
READ TABLE it INTO wa WITH TABLE KEY sec_key COMPONENTS c = 3 d = 4.
"Alias
READ TABLE it INTO wa WITH TABLE KEY sk COMPONENTS c = 3 d = 4.
"Reading a line based on keys specified in a work area
"Work area containing primary and secondary table key values; the line type
"must be compatible to the internal table
DATA(pr_keys) = VALUE struc( a = 1 b = 2 ).
DATA(sec_keys) = VALUE struc( c = 3 d = 4 ).
READ TABLE it FROM pr_keys INTO wa.
"If USING KEY is not specified, the primary table key is used.
"If it is used, the specified table key is used.
READ TABLE it FROM pr_keys USING KEY primary_key INTO wa.
READ TABLE it FROM sec_keys USING KEY sec_key INTO wa.
"Alias
READ TABLE it FROM sec_keys USING KEY sk INTO wa. |
Table expressions |
Find more information about syntax related to table expressions further down. line = it[ TABLE KEY primary_key COMPONENTS a = 1 b = 2 ].
"The addition COMPONENTS is optional; same as above
line = it[ TABLE KEY primary_key a = 1 b = 2 ].
"Primary key alias
line = it[ TABLE KEY pk a = 1 b = 2 ].
"Secondary table key
line = it[ TABLE KEY sec_key c = 3 d = 4 ].
"Secondary table key alias
line = it[ TABLE KEY sk c = 3 d = 4 ].
"Note the section further down about just using KEY (instead of TABLE KEY). |
The specified components used as keys need not be part of a table key.
Statement | Details/Code Snippet |
READ TABLE |
"Note: Table keys are specified with the ... WITH TABLE KEY ... addition,
"free keys with ... WITH KEY ....
READ TABLE it INTO wa WITH KEY b = 2. |
Table expressions |
line = it[ b = 2 ]. |
When reading single lines in general, you can also address individual
components of the line using the component
selector
-
(or the object component selector ->
or the dereferencing
operator
->*
in the case of data reference variables).
You can also use ASSIGN
statements to assign components (and more) to field symbols. See the Example: Exploring READ TABLE Statements and Table Expressions below.
DATA(comp1) = it[ b = 2 ]-c.
READ TABLE it INTO DATA(wa) WITH KEY b = 2.
DATA(comp2) = wa-c.
READ TABLE it ASSIGNING FIELD-SYMBOL(<fs>) WITH KEY b = 2.
DATA(comp3) = <fs>-c.
READ TABLE it REFERENCE INTO DATA(dref) WITH KEY b = 2.
DATA(comp4) = dref->c.
"It is also possible to specify the dereferencing operator
"with the component selector.
DATA(comp5) = dref->*-c.
"Note: When using field symbols and data reference variables as target areas,
"and you modify their content (the examples above only show reads), you can directly
"edit the line in the internal table (because the field symbols/data reference
"variable points to the line). This does not apply to work areas that creates
"local copies. However, you can modify the content of the work area and then,
"for example, use a MODIFY statement to edit the line in the internal table as
"shown further down.
- For example, for checking if a line is found (
sy-subrc
) and stored in the target area, and what the index of the line is (sy-tabix
). - Find more information here
TYPES: BEGIN OF demo_struc,
comp1 TYPE i,
comp2 TYPE c LENGTH 3,
END OF demo_struc,
ty_tab_demo TYPE TABLE OF demo_struc WITH EMPTY KEY.
DATA(it) = VALUE ty_tab_demo( ( comp1 = 1 comp2 = `abc` )
( comp1 = 2 comp2 = `def` )
( comp1 = 1 comp2 = `ghi` ) ).
READ TABLE it INTO DATA(wa) WITH KEY comp2 = `def`.
IF sy-subrc = 0.
... "line found (which is the case in the example)
ELSE.
... "line not found
ENDIF.
ASSERT sy-tabix = 2.
READ TABLE it INTO wa WITH KEY comp2 = `xyz`.
ASSERT sy-subrc = 4.
ASSERT sy-tabix = 0.
- The syntax
READ TABLE ... WHERE ...
searches for the first line that matches aWHERE
condition. - The value of
sy-tabix
is set when a line is found, whilesy-subrc
is set to either 0 or 4. - The
WHERE
condition can use comparison and predicate expressions (see the code snippet comments below for examples). - Syntax options:
- The read result must be specified before the
WHERE
condition. - The
TRANSPORTING NO FIELDS
andUSING KEY
additions are possible. - Dynamic
WHERE
conditions are possible.
- The read result must be specified before the
- If a
READ TABLE
statement can be expressed usingWITH KEY
, this approach - instead of using aWHERE
condition - is recommended as it is more performant. A syntax warning occurs but can be suppressed with the pragma##read_where_ok
. - For optimized searches with a
WHERE
condition:- No optimized search in standard tables without a secondary table key.
- For sorted and hashed keys (sorted/hashed tables, secondary table keys), the search is optimized as described here, if specifications are transferable to a key access.
"Creating and populating demo internal tables
TYPES: BEGIN OF s_demo,
comp1 TYPE i,
comp2 TYPE c LENGTH 5,
comp3 TYPE i,
comp4 TYPE c LENGTH 5,
END OF s_demo,
t_type_so TYPE SORTED TABLE OF s_demo WITH UNIQUE KEY comp1 WITH NON-UNIQUE SORTED KEY sk COMPONENTS comp3.
DATA(itab) = VALUE t_type_so( ( comp1 = 1 comp2 = 'lorem' comp3 = 30 comp4 = 'ipsum' )
( comp1 = 2 comp2 = 'dolor' comp3 = 20 comp4 = 'sit' )
( comp1 = 3 comp2 = 'amet' comp3 = 40 comp4 = 'hello' )
( comp1 = 4 comp2 = 'world' comp3 = 50 comp4 = 'ABAP' )
( comp1 = 5 comp2 = 'test' comp3 = 10 comp4 = '' ) ).
DATA wa TYPE s_demo.
"--------- WHERE condition with comparison expressions ---------
"Examples: =/EQ, <>/NE, >/GT, </LT, >=,GE, <=/LE,
" CO, CN, CA, NA, CS, NS, CP, NP,
" [NOT] BETWEEN ... AND
" [NOT] IN ranges_tables
READ TABLE itab INTO wa WHERE comp1 > 3.
ASSERT sy-tabix = 4.
"'or' also available in other lines in this component, but the first found line
"is returned (lines 1, 2, 4)
READ TABLE itab INTO wa WHERE comp2 CS 'or'.
ASSERT sy-tabix = 1.
"'d' occurs in lines 2, 4
READ TABLE itab INTO wa WHERE comp2 CS 'd'.
ASSERT sy-tabix = 2.
READ TABLE itab INTO wa WHERE comp2 CS 'd'.
ASSERT sy-tabix = 2.
"--------- WHERE condition with predicate expressions ---------
"Examples: IS [NOT] INITIAL
" IS [NOT] BOUND
" IS [NOT] INSTANCE OF
READ TABLE itab INTO wa WHERE comp1 > 4 AND comp4 IS INITIAL.
ASSERT sy-tabix = 5.
"--------- WITH KEY instead of WHERE condition ---------
"Syntax warning in READ TABLE ... WHERE ... statements
READ TABLE itab INTO wa WHERE comp4 IS INITIAL.
ASSERT sy-tabix = 5.
"For a better performance, the previous statement should be
"replaced by a READ TABLE ... WITH KEY ... statement.
READ TABLE itab INTO wa WITH KEY comp4 = ''.
ASSERT sy-tabix = 5.
"You can also suppress the syntax warning by a pragma.
READ TABLE itab INTO wa WHERE comp4 IS INITIAL ##read_where_ok.
ASSERT sy-tabix = 5.
"------------------- Further additions -------------------
"TRANSPORTING NO FIELDS addition is possible
READ TABLE itab TRANSPORTING NO FIELDS WHERE comp2 CS 'd'.
ASSERT sy-tabix = 2.
"USING KEY addition
READ TABLE itab USING KEY primary_key INTO wa WHERE comp2 CS 't'.
ASSERT sy-tabix = 3.
READ TABLE itab USING KEY sk INTO wa WHERE comp3 > 40.
ASSERT sy-tabix = 5.
"------------------- Excursions -------------------
"Note the comparison rules for character-like data types
READ TABLE itab INTO wa WHERE comp2 = 'lorem' ##read_where_ok.
ASSERT sy-tabix = 1.
"In the following case, the length of the comp2 value increased to match the
"length of the specified text field, i.e. the surplus characters from the right
"side text field are not truncated. As a consequence, the line is not found.
DATA(some_text) = 'loremXYZ'.
READ TABLE itab INTO wa WHERE comp2 = some_text ##read_where_ok.
ASSERT sy-tabix = 0 AND sy-subrc <> 0.
"When using READ TABLE ... WITH KEY ... the behavior is different. In that
"case, the surplus characters are truncated because of a conversion. Therefore,
"the following statement finds a line.
READ TABLE itab INTO wa WITH KEY comp2 = some_text.
ASSERT sy-tabix = 1 AND sy-subrc = 0.
"Note: The read target can only be placed before WHERE conditions.
"The following statements are not possible.
"READ TABLE itab WHERE comp2 CS 'd' INTO wa.
"READ TABLE itab WHERE comp2 CS 'd' TRANSPORTING NO FIELDS.
"READ TABLE ... WHERE ... statements can replace LOOP AT ... WHERE ...
"statements including EXIT.
LOOP AT itab INTO wa WHERE comp2 CS 'd'.
ASSERT sy-tabix = 2.
EXIT.
ENDLOOP.
"------------------- Dynamic WHERE condition -------------------
"Character-like data objects or standard tables with character-like line type
"can be specified
DATA(dyn_where_cond_str) = `comp2 CS 'd'`.
READ TABLE itab INTO wa WHERE (dyn_where_cond_str).
ASSERT sy-tabix = 2.
DATA(dyn_where_cond_tab) = VALUE string_table( ( `comp2` ) ( `CS` ) ( `'d'` ) ).
READ TABLE itab INTO wa WHERE (dyn_where_cond_tab).
ASSERT sy-tabix = 2.
... TRANSPORTING NO FIELDS
: It is only checked whether the line exists. No target area is specified. As mentioned above, the system fields are filled.
DATA(stringtab) = VALUE string_table( ( `abc` ) ( `def` ) ( `ghi` ) ).
READ TABLE stringtab WITH KEY table_line = `def` TRANSPORTING NO FIELDS.
ASSERT sy-subrc = 0.
ASSERT sy-tabix = 2.
READ TABLE stringtab WITH KEY table_line = `xyz` TRANSPORTING NO FIELDS.
ASSERT sy-subrc = 4.
ASSERT sy-tabix = 0.
... TRANSPORTING ...
: Specifying fields to be transported. The addition cannot be used with the ASSIGNING
and REFERENCE
additions.
TYPES: BEGIN OF demo_struc_tr,
comp1 TYPE i,
comp2 TYPE c LENGTH 3,
END OF demo_struc_tr,
ty_tab_demo_tr TYPE TABLE OF demo_struc_tr WITH EMPTY KEY.
DATA(it_tr) = VALUE ty_tab_demo_tr( ( comp1 = 1 comp2 = `abc` )
( comp1 = 2 comp2 = `def` ) ).
READ TABLE it_tr INTO DATA(line_tr) INDEX 1 TRANSPORTING comp2.
ASSERT line_tr-comp1 IS INITIAL.
"ALL FIELDS addition is available, explicitly stating that all fields
"should be transported.
READ TABLE it_tr INTO line_tr INDEX 1 TRANSPORTING ALL FIELDS.
... COMPARING ...
:
- Can be used together with and in front of
TRANSPORTING ...
- Compares specified components
- The
COMPARING ALL FIELDS
compares all components,COMPARING NO FIELDS
compares no components - Setting of
sy-subrc
: 0 is set if the content of compared components is identical, otherwise it is 2. Found lines are nevertheless assigned independently of the comparison.
TYPES: BEGIN OF demo_struc_compare,
comp1 TYPE i,
comp2 TYPE c LENGTH 3,
comp3 TYPE string,
END OF demo_struc_compare,
ty_tab_demo_compare TYPE TABLE OF demo_struc_compare WITH EMPTY KEY.
DATA(it_compare) = VALUE ty_tab_demo_compare( ( comp1 = 1 comp2 = 'abc' comp3 = `zzz` )
( comp1 = 2 comp2 = 'def' comp3 = `yyy` )
( comp1 = 2 comp2 = 'ghi' comp3 = `yyy` ) ).
DO 3 TIMES.
DATA(line_compare) = VALUE demo_struc_compare( comp1 = 1 ).
READ TABLE it_compare INTO line_compare INDEX sy-index COMPARING comp1.
CASE sy-index.
WHEN 1.
ASSERT sy-subrc = 0.
WHEN 2.
ASSERT sy-subrc = 2.
WHEN 3.
ASSERT sy-subrc = 2.
ENDCASE.
line_compare = VALUE demo_struc_compare( comp3 = `yyy` ).
READ TABLE it_compare INTO line_compare INDEX sy-index COMPARING comp3 TRANSPORTING comp2 comp3.
CASE sy-index.
WHEN 1.
ASSERT sy-subrc = 2.
ASSERT line_compare-comp1 IS INITIAL.
WHEN 2.
ASSERT sy-subrc = 0.
ASSERT line_compare-comp1 IS INITIAL.
WHEN 3.
ASSERT sy-subrc = 0.
ASSERT line_compare-comp1 IS INITIAL.
ENDCASE.
ENDDO.
TYPES c3 TYPE c LENGTH 3.
TYPES ty_tab_casting TYPE TABLE OF c3 WITH EMPTY KEY.
DATA(it_casting) = VALUE ty_tab_casting( ( 'abc' ) ( 'def' ) ).
"Field symbol created inline (i.e. with the generic type 'data' implicitly)
"In this case, the CASTING and ELSE UNASSIGN additions are not available.
READ TABLE it_casting ASSIGNING FIELD-SYMBOL(<a>) INDEX 1.
"******* CASTING addition *******
"To use the addition, the field symbol must be either completely typed, or
"typed with one of the generic built-in ABAP types c, n, p, or x.
TYPES c2 TYPE c LENGTH 2.
FIELD-SYMBOLS <b> TYPE c2.
READ TABLE it_casting ASSIGNING <b> CASTING INDEX 2.
ASSERT <b> = 'de'.
"******* ELSE UNASSIGN addition *******
"The field symbol is unassigned if no table line is found. The addition
"can be used together with the CASTING addition.
"The following example loops 3 times across an internal table that has
"two lines. The sy-index value is used as the index value of the READ TABLE
"statement. The example demonstrates that when a line is not found, the field
"symbol is unassigned. In case of the first READ TABLE statement that does not
"specify ELSE UNASSIGN, the field symbol remains assigned.
FIELD-SYMBOLS <c> TYPE c2.
DO 3 TIMES.
READ TABLE it_casting ASSIGNING FIELD-SYMBOL(<d>) INDEX sy-index.
READ TABLE it_casting ASSIGNING FIELD-SYMBOL(<e>) ELSE UNASSIGN INDEX sy-index.
READ TABLE it_casting ASSIGNING <c> CASTING ELSE UNASSIGN INDEX sy-index.
IF sy-index = 3.
ASSERT <d> = `def`.
ASSERT <e> IS NOT ASSIGNED.
ASSERT <c> IS NOT ASSIGNED.
ENDIF.
ENDDO.
You may stumble on the BINARY SEARCH
addition to READ TABLE
statements for an optimized access. However, it is recommended to use sorted tables or secondary table keys for an optimized access. Find information and an example in the expandable section below.
READ TABLE itab WITH KEY ... BINARY SEARCH ...
🟢 Click to expand for more information and a code example
READ TABLE
withoutBINARY SEARCH
: Table is accessed linearlyREAD TABLE
withBINARY SEARCH
: Table is accessed using a binary search- Using the
BINARY SEARCH
addition is particularly more efficient for larger tables when accessing data often. - The table must be sorted in ascending order based on the keys being searched.
BINARY SEARCH
is suitable for standard tables that do not have a secondary key defined and when you need to make multiple read accesses to the table (however, note the costs of a previous sorting)BINARY SEARCH
can only be used with index tables and not with hashed tables. If the table is a sorted table and the read access uses a free key, the addition can only be applied when the initial part of the table key is specified. I.e. if key components area
,b
, andc
, the addition can be used by specifyinga
alone,a
andb
, ora
,b
, andc
. However, it means that the syntax can be specified without errors. TheBINARY SEARCH
specifiation has no effect and is redundant. Syntactically not possible withBINARY SEARCH
(for example):b
andc
withouta
, or any other non-key component (because it cannot be sorted according to the non-key component).
- Using the
- Depending on the number of times you need to access the internal table, it is recommended to work with sorted tables or tables with secondary keys. If you only need to read one or a few data sets, consider the administrative costs of setting up the index.
- Note: The
BINARY SEARCH
addition is not available for table expressions. IfKEY ...
is specified, an optimized search is performed by default. There are no performance differences between using theREAD TABLE
statement and table expressions.
To try it out the following example, create a demo class named zcl_some_class
and paste the code into it. After activation, choose F9 in ADT to execute the class. The example is set up to display output in the console.
The example includes multiple reads on standard internal tables using a READ TABLE
statement ...
- without
BINARY SEARCH
. - with
BINARY SEARCH
and a previousSORT
statement. - with a secondary table key whose components correspond to free key used in the previous statements.
The runtime of the reads is determined and stored in internal tables. The read operations are repeated several times to have a more accurate runtime evaluation. The fastest time is output.
The example is intended to demonstrate the performance gain using the
BINARY SEARCH
addition (and also using a secondary table key).
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
"Line type and internal table declarations
TYPES: BEGIN OF demo_struc,
idx TYPE i,
str TYPE string,
num TYPE i,
END OF demo_struc.
"Tables with empty primary table key
DATA itab_std1 TYPE STANDARD TABLE OF demo_struc WITH EMPTY KEY.
DATA itab_std2 LIKE itab_std1.
"Table with empty primary table key, secondary table key specified
DATA itab_sec TYPE STANDARD TABLE OF demo_struc
WITH EMPTY KEY
WITH NON-UNIQUE SORTED KEY sk COMPONENTS str num.
DATA num_of_table_lines TYPE i VALUE 5000.
DATA num_of_repetitions TYPE i VALUE 10.
DATA num_of_reads TYPE i VALUE 3000.
"Populating internal tables
DO num_of_table_lines TIMES.
INSERT VALUE #( idx = sy-index
str = |INDEX{ sy-index }|
num = sy-index ) INTO TABLE itab_std1.
ENDDO.
itab_std2 = itab_std1.
itab_sec = itab_std1.
DATA no_binary_search TYPE TABLE OF decfloat34 WITH EMPTY KEY.
DATA with_sort_and_binary_search TYPE TABLE OF decfloat34 WITH EMPTY KEY.
DATA with_secondary_key TYPE TABLE OF decfloat34 WITH EMPTY KEY.
"Repeating the reads several times for a more accurate result
DO num_of_repetitions TIMES.
"---- Reading without the BINARY SEARCH addition ----
DATA(ts1) = utclong_current( ).
DO num_of_reads TIMES.
READ TABLE itab_std1 WITH KEY str = `INDEX` && sy-index num = sy-index TRANSPORTING NO FIELDS.
ENDDO.
DATA(ts2) = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = DATA(seconds) ).
APPEND seconds TO no_binary_search.
"---- Reading with the BINARY SEARCH addition ----
ts1 = utclong_current( ).
"Sorting the internal table when using BINARY SEARCH
"In this simple example, the internal table is populated by having the free key components
"to be searched in ascending order anyway. This is to emphasize the requirement to
"sort the (standard) internal table when using BINARY SEARCH. Here, the SORT statement
"is counted to the runtime.
SORT itab_std2 BY str num.
DO num_of_reads TIMES.
READ TABLE itab_std2 WITH KEY str = `INDEX` && sy-index num = sy-index BINARY SEARCH TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
APPEND seconds TO with_sort_and_binary_search.
"---- Excursion: Reading with READ TABLE using a secondary table key ----
ts1 = utclong_current( ).
DO num_of_reads TIMES.
READ TABLE itab_sec WITH TABLE KEY sk COMPONENTS str = `INDEX` && sy-index num = sy-index TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
APPEND seconds TO with_secondary_key.
ENDDO.
SORT no_binary_search ASCENDING BY table_line.
SORT with_sort_and_binary_search ASCENDING BY table_line.
SORT with_secondary_key ASCENDING BY table_line.
out->write( |Number of read repetitions: { num_of_repetitions }| ).
out->write( |Number of reads per table: { num_of_reads }\n| ).
out->write( `Fastest run of reads using READ TABLE with a free key, without the BINARY SEARCH addition:` ).
out->write( no_binary_search[ 1 ] ).
out->write( repeat( val = `-` occ = 70 ) ).
out->write( `Fastest run of reads using SORT, and READ TABLE with a free key, and the BINARY SEARCH addition:` ).
out->write( with_sort_and_binary_search[ 1 ] ).
out->write( repeat( val = `-` occ = 70 ) ).
out->write( `Fastest run of reads using READ TABLE and a secondary table key:` ).
out->write( with_secondary_key[ 1 ] ).
ENDMETHOD.
ENDCLASS.
Expand the following collapsible section for example code. To try it out, create a demo class named zcl_some_class
and paste the code into it. After activation, choose F9 in ADT to execute the class.
The example is not set up to display output in the console. You may want to set a break point at the first position possible and walk through the example in the debugger. This will allow you to double-click on data objects and observe how the different statements affect their contents.
🟢 Click to expand for example code
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
"Creating and populating an internal table
TYPES: BEGIN OF st_h,
a TYPE i,
b TYPE c LENGTH 2,
c TYPE string,
END OF st_h,
ty_tab_f TYPE SORTED TABLE OF st_h WITH UNIQUE KEY a WITH NON-UNIQUE SORTED KEY sk COMPONENTS b.
DATA(it_j) = VALUE ty_tab_f( ( a = 1 b = 'zz' c = `B` )
( a = 2 b = 'xx' c = `D` )
( a = 3 b = 'yy' c = `F` ) ).
"******* 1) Reading a single line by index: READ TABLE statements *******
DATA struc_e TYPE st_h.
"In the following example ...
"- a work area is specified as target area.
"- USING KEY is not specified, i.e. the primary table index is used by default.
READ TABLE it_j INTO struc_e INDEX 2.
"Reading into a work area that is created inline
READ TABLE it_j INTO DATA(struc_f) INDEX 3.
"Specifying other target areas: Field symbols and data reference variables
"Here, the target areas are created inline
READ TABLE it_j ASSIGNING FIELD-SYMBOL(<fs_c>) INDEX 1.
READ TABLE it_j REFERENCE INTO DATA(dref_c) INDEX 1.
"******* USING KEY addition *******
"Reading by index and specifying which table index to use
"In the following example, the primary key is specified explicitly and
"addressed using the default name primary_key. It has the same effect
"as the statement below because the primary table index is used by
"default.
READ TABLE it_j INTO struc_e INDEX 1 USING KEY primary_key.
READ TABLE it_j INTO struc_e INDEX 1.
"Specifying the secondary key to use the secondary table index
READ TABLE it_j INTO struc_e INDEX 1 USING KEY sk.
"Using alias names
DATA it_j_alias TYPE SORTED TABLE OF st_h
WITH UNIQUE KEY primary_key ALIAS pk COMPONENTS a
WITH NON-UNIQUE SORTED KEY sk ALIAS sk_alias COMPONENTS b.
it_j_alias = it_j.
READ TABLE it_j_alias INTO struc_e INDEX 1 USING KEY pk.
READ TABLE it_j_alias INTO struc_e INDEX 1 USING KEY sk_alias.
"The following examples use the other key names
READ TABLE it_j_alias INTO struc_e INDEX 1 USING KEY primary_key.
READ TABLE it_j_alias INTO struc_e INDEX 1 USING KEY sk.
"******* Excursion: System field setting with READ TABLE statements *******
"Checking if a line is found and stored in the target area
READ TABLE it_j INTO struc_e INDEX 999.
IF sy-subrc = 0.
...
ELSE.
... "This branch is executed in the example since the line is not found.
ASSERT sy-tabix = 0.
ENDIF.
READ TABLE it_j INTO struc_e INDEX 1.
ASSERT sy-subrc = 0.
ASSERT sy-tabix = 1.
"******* 2) Reading a single line by index: Table expressions *******
"Many of the following examples show the assignment of a line
"using a table expression that is specified on the right side
"of an assignment.
"The target area is declared inline on the left side of an assignment.
DATA(struc_g) = it_j[ 1 ].
"If a table line is not found, a catchable exception is raised. It can
"be caught in a TRY control structure.
TRY.
struc_g = it_j[ 999 ].
CATCH cx_sy_itab_line_not_found.
ENDTRY.
"Table expression on the left side of an assignment
"The following example does not work since the example is a key table.
"it_j[ 1 ] = VALUE #( ).
"Creating and populating a table without a key
DATA it_j_no_key TYPE TABLE OF st_h WITH EMPTY KEY.
it_j_no_key = VALUE #( ( a = 1 b = 'a' c = `b` )
( a = 2 b = 'c' c = `d` )
( a = 3 b = 'e' c = `f` ) ).
it_j_no_key[ 1 ] = VALUE #( b = 'z' ).
ASSERT it_j_no_key[ 1 ]-a IS INITIAL.
ASSERT it_j_no_key[ 1 ]-b = 'z'.
ASSERT it_j_no_key[ 1 ]-c IS INITIAL.
it_j_no_key[ 2 ]-b = 'x'.
it_j_no_key[ 3 ] = VALUE #( ).
ASSERT it_j_no_key[ 3 ] IS INITIAL.
"******* KEY addition *******
"Reading by index and specifying which table index to use
"Primary key's default name
struc_g = it_j_alias[ KEY primary_key INDEX 1 ].
"Primary key alias
struc_g = it_j_alias[ KEY pk INDEX 1 ].
"Secondary key
struc_g = it_j_alias[ KEY sk INDEX 1 ].
"Secondary key alias
struc_g = it_j_alias[ KEY sk_alias INDEX 1 ].
"******* Excursion: Other ways of assignments with table expressions *******
"The examples only read by index. Reading by keys is also possible.
"Assigning a line read by index to a field symbol
FIELD-SYMBOLS <fs_d> TYPE st_h.
ASSIGN it_j[ 1 ] TO <fs_d>.
"The field symbol created inline has a generic type
ASSIGN it_j[ 1 ] TO FIELD-SYMBOL(<fs_d_inl>).
"Copying a table line via a table expression and embedding it in a
"constructor expression
struc_g = VALUE #( it_j[ 3 ] ).
"Reading into data reference variable using the REF operator
DATA(dref_d) = REF #( it_j[ 2 ] ).
"******* OPTIONAL/DEFAULT additions *******
"OPTIONAL: The default value is an initial data object with the data type
DATA(struc_h) = VALUE #( it_j[ 999 ] OPTIONAL ).
ASSERT struc_h = VALUE st_h( ).
"DEFAULT: Specifying a default value for table lines not found
"This value must be convertible to the type of the table expression.
"It can also be a table expression.
struc_h = VALUE #( it_j[ 999 ] DEFAULT it_j[ 1 ] ).
struc_h = VALUE #( it_j[ 999 ] DEFAULT VALUE st_h( c = `null` ) ).
**********************************************************************
"******* 1) Reading a single line using table keys: READ TABLE statements *******
"Creating and populating a demo internal table
TYPES: BEGIN OF st_i,
num TYPE i,
str TYPE string,
char TYPE c LENGTH 2,
END OF st_i.
DATA it_k TYPE SORTED TABLE OF st_i
WITH NON-UNIQUE KEY primary_key ALIAS pk COMPONENTS num
WITH NON-UNIQUE SORTED KEY sec_key ALIAS sk COMPONENTS char.
it_k = VALUE #( ( num = 1 str = `A` char = 'zz' )
( num = 2 str = `C` char = 'yy' )
( num = 3 str = `E` char = 'xx' ) ).
"The following examples use a work area as target. Other target areas are
"possible.
"Primary table key
READ TABLE it_k INTO DATA(struc_i) WITH TABLE KEY primary_key COMPONENTS num = 3.
"Primary table key alias
READ TABLE it_k INTO struc_i WITH TABLE KEY pk COMPONENTS num = 2.
"Secondary table key
READ TABLE it_k INTO struc_i WITH TABLE KEY sec_key COMPONENTS char = 'xx'.
"Secondary table key alias
READ TABLE it_k INTO struc_i WITH TABLE KEY sk COMPONENTS char = 'yy'.
"Reading a line based on keys specified in a work area
"It is a work area containing primary and secondary table key values.
"the line type must be compatible to the internal table.
TYPES st_j LIKE LINE OF it_k.
DATA(pr_key) = VALUE st_j( num = 1 ).
DATA(sec_key) = VALUE st_j( char = 'yy' ).
READ TABLE it_k FROM pr_key INTO struc_i.
"If USING KEY is not specified, the primary table key is used by default.
"Explicitly specifying the primary table key
READ TABLE it_k FROM pr_key USING KEY primary_key INTO struc_i.
"Primary table key alias
READ TABLE it_k FROM pr_key USING KEY pk INTO struc_i.
"Secondary table key
READ TABLE it_k FROM sec_key USING KEY sec_key INTO struc_i.
"Secondary table key alias
READ TABLE it_k FROM sec_key USING KEY sk INTO struc_i.
"******* 2) Reading a single line using table keys: Table expressions *******
"The addition COMPONENTS is optional. Therefore, the statement below
"has the same effect.
"Explicitly specifying the primary table key
struc_i = it_k[ KEY primary_key COMPONENTS num = 1 ].
struc_i = it_k[ KEY primary_key num = 1 ].
"Primary table key alias
struc_i = it_k[ KEY pk num = 2 ].
"Secondary table key
struc_i = it_k[ KEY sec_key char = 'xx' ].
"Secondary table key alias
struc_i = it_k[ KEY sk char = 'zz' ].
**********************************************************************
"Reading a single line using a free key
"Note: Instead if READ TABLE ... WITH TABLE KEY ..., it is ... WITH KEY.
READ TABLE it_k INTO DATA(struc_j) WITH KEY str = `A`.
struc_j = it_k[ str = `C` ].
**********************************************************************
"Examples for addressing individual components of read lines
"The assertions emphasize the difference of work areas and field
"symbols/data reference variables as target areas. Modifying the
"contents of the field symbols/data reference variables means
"modifying the internal table content.
DATA(comp_a) = it_k[ str = `C` ]-num.
READ TABLE it_k INTO DATA(struc_k) WITH KEY str = `A`.
struc_k-num = 123.
ASSERT NOT it_k[ str = `C` ]-num = 123.
DATA(comp_b) = struc_k-num.
READ TABLE it_k ASSIGNING FIELD-SYMBOL(<fs_e>) WITH KEY str = `C`.
"Note: The example table is a sorted table with 'num' as part of
"a unique key. The field value cannot be modified.
"<fs_e>-num = 123.
<fs_e>-char = 'hi'.
ASSERT it_k[ str = `C` ]-char = 'hi'.
DATA(comp_c) = <fs_e>-char.
READ TABLE it_k REFERENCE INTO DATA(dref_e) WITH KEY str = `E`.
dref_e->char = '##'.
ASSERT it_k[ str = `E` ]-char = '##'.
DATA(comp_d) = dref_e->num.
"It is also possible to specify the dereferencing operator together
"with the component selector.
DATA(comp_e) = dref_e->*-char.
**********************************************************************
"Excursion: Using ASSIGN statements to assign to field symbols
TYPES ty_tab_h TYPE TABLE OF st_i WITH EMPTY KEY.
DATA(it_l) = VALUE ty_tab_h( ( num = 1 str = `A` char = 'zz' )
( num = 2 str = `C` char = 'yy' )
( num = 3 str = `E` char = 'xx' ) ).
FIELD-SYMBOLS <fs_f> TYPE st_i-num.
ASSIGN it_l[ str = `C` ]-num TO <fs_f>.
<fs_f> = 99.
ASSERT it_l[ str = `C` ]-num = 99.
FIELD-SYMBOLS <fs_g> TYPE st_i-char.
ASSIGN it_l[ str = `C` ]-char TO <fs_g>.
<fs_g> = 'aa'.
"Field symbol created inline
ASSIGN <fs_g> TO FIELD-SYMBOL(<fs_h>).
<fs_h> = 'bb'.
READ TABLE it_l REFERENCE INTO DATA(dref_f) WITH KEY str = `E`.
ASSIGN dref_f->char TO FIELD-SYMBOL(<fs_i>).
<fs_i> = 'cc'.
"The following example use the generic type any so that anything
"can be assigned.
FIELD-SYMBOLS <fs_j> TYPE any.
"Assigning component
ASSIGN it_l[ str = `C` ]-num TO <fs_j>.
<fs_j> = 123.
"Assigning line
ASSIGN it_l[ str = `C` ] TO <fs_j>.
<fs_j> = VALUE st_i( num = 789 str = `U` char = 'vv' ).
"Assigning the entire table
ASSIGN it_l TO <fs_j>.
<fs_j> = VALUE ty_tab_h( ( num = 1 str = `AB` char = 'ap' ) ).
**********************************************************************
"COMPARING / TRANSPORTING additions to READ TABLE statements
"Comparing fields and specifying fields to be transported
DATA(it_m) = VALUE ty_tab_h( ( num = 1 str = `Z` char = '##' )
( num = 2 str = `Y` char = 'yy' )
( num = 3 str = `X` char = '##' )
( num = 4 str = `W` char = 'ww' )
( num = 5 str = `W` char = '##' )
( num = 6 str = `V` char = '##' )
( num = 7 str = `V` char = '##' )
( num = 7 str = `V` char = '##' )
( num = 8 str = `V` char = 'vv' ) ).
"******* TRANSPORTING NO FIELDS addition *******
"It is only checked whether the line exists. No target area is specified.
"The system fields sy-subrc and sy-tabix are filled. Check also the
"line_exists and line_index functions.
READ TABLE it_m WITH KEY str = `X` TRANSPORTING NO FIELDS.
ASSERT sy-subrc = 0.
DATA(sysubrc) = sy-subrc.
ASSERT sy-tabix = 3.
DATA(sytabix) = sy-tabix.
READ TABLE it_m WITH KEY str = `nope` TRANSPORTING NO FIELDS.
ASSERT sy-subrc = 4.
ASSERT sy-tabix = 0.
"******* TRANSPORTING ... addition *******
"Specifying fields to be transported; cannot be used with the ASSIGNING
"and REFERENCE additions
READ TABLE it_m INTO DATA(struc_l) INDEX 1 TRANSPORTING num char.
ASSERT struc_l-str IS INITIAL.
"If ALL FIELDS is specified, all fields are assigned, which corresponds to the
"example below.
READ TABLE it_m INTO struc_l INDEX 1 TRANSPORTING ALL FIELDS.
READ TABLE it_m INTO struc_l INDEX 1.
"******* COMPARING addition *******
"- Can be used together with and in front of TRANSPORTING ...
"- Compares the specified components
"- ALL FIELDS compares all components, NO FIELDS compares no components
"- Setting of sy-subrc: 0 is set if the content of compared components is identical,
" otherwise it is 2. Found lines are nevertheless assigned independently of the comparison.
"The following examples use a WHILE loop to read all table lines (sy-index represents the
"index value of the primary table index) into a work area.
"The work area is filled before the read for the comparison. Depending on the comparison
"result (by checking the sy-subrc value), the lines are added to different internal tables
"for demonstration purposes. In addition, the 'num' component value is added to a string.
"The examples explore several syntax options.
DATA struc_m LIKE LINE OF it_m.
DATA it_n LIKE it_m.
DATA it_o LIKE it_m.
DATA nums_subrc_0 TYPE string.
DATA nums_subrc_2 TYPE string.
DATA(subrc) = 0.
"Specifying ALL FIELDS
WHILE subrc = 0.
DATA(idx) = sy-index.
struc_m = VALUE #( num = 7 str = `V` char = '##' ).
READ TABLE it_m INTO struc_m INDEX idx COMPARING ALL FIELDS TRANSPORTING ALL FIELDS.
subrc = COND #( WHEN sy-subrc = 0 THEN 0 ELSE sy-subrc ).
IF subrc = 0.
APPEND struc_m TO it_n.
nums_subrc_0 &&= struc_m-num.
ELSEIF subrc = 2.
APPEND struc_m TO it_o.
nums_subrc_2 &&= struc_m-num.
subrc = 0.
ELSE.
EXIT.
ENDIF.
ENDWHILE.
ASSERT nums_subrc_0 = `77`.
ASSERT nums_subrc_2 = `1234568`.
CLEAR: subrc, struc_m, it_n, it_o, nums_subrc_0, nums_subrc_2.
"Specifying specific fields for the comparison and transport
WHILE subrc = 0.
idx = sy-index.
struc_m = VALUE #( num = 1234 str = `NOPE` char = '##' ).
READ TABLE it_m INTO struc_m INDEX idx COMPARING char TRANSPORTING num.
subrc = COND #( WHEN sy-subrc = 0 THEN 0 ELSE sy-subrc ).
IF subrc = 0.
APPEND struc_m TO it_n.
nums_subrc_0 &&= struc_m-num.
ELSEIF subrc = 2.
APPEND struc_m TO it_o.
nums_subrc_2 &&= struc_m-num.
subrc = 0.
ELSE.
EXIT.
ENDIF.
ENDWHILE.
ASSERT nums_subrc_0 = `135677`.
ASSERT nums_subrc_2 = `248`.
CLEAR: subrc, struc_m, it_n, it_o, nums_subrc_0, nums_subrc_2.
WHILE subrc = 0.
idx = sy-index.
struc_m = VALUE #( num = 9999 char = '##' str = `V` ).
READ TABLE it_m INTO struc_m INDEX idx COMPARING char str TRANSPORTING num.
subrc = COND #( WHEN sy-subrc = 0 THEN 0 ELSE sy-subrc ).
IF subrc = 0.
APPEND struc_m TO it_n.
nums_subrc_0 &&= struc_m-num.
ELSEIF subrc = 2.
APPEND struc_m TO it_o.
nums_subrc_2 &&= struc_m-num.
subrc = 0.
ELSE.
EXIT.
ENDIF.
ENDWHILE.
ASSERT nums_subrc_0 = `677`.
ASSERT nums_subrc_2 = `123458`.
**********************************************************************
"CASTING / ELSE UNASSIGN additions to READ TABLE statements
"Additions when assigning the read result to a field symbol
TYPES c3 TYPE c LENGTH 3.
TYPES ty_tab_g TYPE TABLE OF c3 WITH EMPTY KEY.
DATA(itq) = VALUE ty_tab_g( ( 'abc' ) ( 'def' ) ).
"Field symbol created inline (i.e. with the generic type 'data' implicitly)
"In this case, the CASTING and ELSE UNASSIGN additions are not available.
READ TABLE itq ASSIGNING FIELD-SYMBOL(<fs_k>) INDEX 1.
"******* CASTING addition *******
"To use the addition, the field symbol must be either completely typed, or
"typed with one of the generic built-in ABAP types c, n, p, or x.
TYPES c2 TYPE c LENGTH 2.
FIELD-SYMBOLS <fs_l> TYPE c2.
READ TABLE itq ASSIGNING <fs_l> CASTING INDEX 2.
ASSERT <fs_l> = 'de'.
"******* ELSE UNASSIGN addition *******
"The field symbol is unassigned if no table line is found. The addition
"can be used together with the CASTING addition.
"The following example loops 3 times across an internal table that has
"two lines. The sy-index value is used as the index value of the READ TABLE
"statement. The example demonstrates that when a line is not found, the field
"symbol is unassigned. In case of the first READ TABLE statement that does not
"specify ELSE UNASSIGN, the field symbol remains assigned.
DATA string_a TYPE string.
FIELD-SYMBOLS <fs_o> TYPE c2.
DO 3 TIMES.
READ TABLE itq ASSIGNING FIELD-SYMBOL(<fs_m>) INDEX sy-index.
READ TABLE itq ASSIGNING FIELD-SYMBOL(<fs_n>) ELSE UNASSIGN INDEX sy-index.
READ TABLE itq ASSIGNING <fs_o> CASTING ELSE UNASSIGN INDEX sy-index.
IF sy-index = 3.
ASSERT <fs_m> = `def`.
ASSERT <fs_n> IS NOT ASSIGNED.
ASSERT <fs_o> IS NOT ASSIGNED.
ENDIF.
ENDDO.
ENDMETHOD.
ENDCLASS.
Table expressions have been mentioned in sections above. This is to summarize.
- Table expressions typically use the name of an internal table followed by square brackets with content
[ ... ]
, specifying a table line. - They allow read and write access to internal tables at various positions.
- These expressions are a concise form of
READ TABLE
statements, enabling read and write operations in operand positions.- Unlike
READ TABLE
, table expressions do not change thesy-tabix
system field (except withASSIGN
).
- Unlike
- After the brackets, you can specify chaining (component and object component selectors or additional square brackets). Without chaining, the entire line is respected.
- Note that not found lines raise the catchable exception
CX_SY_ITAB_LINE_NOT_FOUND
.
The following table demonstrates a selection of subjects where table expressions are applicable.
Subject | Details/Code Snippet |
Reading table lines by index |
"Creating and populating demo internal tables
"Note: These demo tables are relevant for most of the code snippets
"in this section.
TYPES: BEGIN OF s_demo,
comp1 TYPE i,
comp2 TYPE i,
comp3 TYPE i,
comp4 TYPE c LENGTH 3,
END OF s_demo,
ttyp TYPE SORTED TABLE OF s_demo WITH UNIQUE KEY comp1 WITH NON-UNIQUE SORTED KEY sk COMPONENTS comp2 comp3,
ttyp_hashed TYPE HASHED TABLE OF s_demo WITH UNIQUE KEY comp1 WITH NON-UNIQUE SORTED KEY sk COMPONENTS comp2 comp3,
ttyp2 TYPE SORTED TABLE OF s_demo WITH UNIQUE KEY comp1 comp2 WITH NON-UNIQUE SORTED KEY sk COMPONENTS comp3.
DATA(itab) = VALUE ttyp( ( comp1 = 1 comp2 = 30 comp3 = 31 comp4 = 'aaa' )
( comp1 = 2 comp2 = 20 comp3 = 21 comp4 = 'bbb' )
( comp1 = 3 comp2 = 10 comp3 = 11 comp4 = 'ccc' ) ).
DATA itab_hashed TYPE ttyp_hashed.
itab_hashed = itab.
DATA itab_so TYPE ttyp2.
itab_so = itab.
DATA line TYPE s_demo.
"------ Reading table line by index------
"Just specifying the index number means referring to the primary table index.
"In this case, the internal table must be an index table.
"In the example, the entire table line is assigned to a variable
line = itab[ 2 ].
"KEY ... INDEX ... additions
"For reading a line according to a table index.
"The following example has the same effect as above. Here, the default
"name of the primary key is specified explicitly.
line = itab[ KEY primary_key INDEX 2 ].
"Secondary table key specified, using secondary table index
line = itab[ KEY sk INDEX 1 ].
"This syntax is not possible for hashed tables.
"DATA(line_hashed_tab1) = itab_hashed[ 2 ].
"DATA(line_hashed_tab2) = itab_hashed[ KEY primary_key INDEX 2 ].
"Secondary table index access is possible for hashed tables
DATA(line_hashed_tab3) = itab_hashed[ KEY sk INDEX 2 ]. |
Reading table lines by table key |
"------------------ TABLE KEY addition ------------------
"Explicitly specifying the primary table key
line = itab[ TABLE KEY primary_key COMPONENTS comp1 = 1 ].
"The following statement is not possible as no other components can be specified.
"line = itab[ TABLE KEY primary_key COMPONENTS comp1 = 1 comp2 = 30 ].
"The addition COMPONENTS is optional; the following example is the same as above
line = itab[ TABLE KEY primary_key comp1 = 1 ].
"Specifying a secondary table key
line = itab[ TABLE KEY sk COMPONENTS comp2 = 20 comp3 = 21 ].
"Optional COMPONENTS addition
line = itab[ TABLE KEY sk comp2 = 20 comp3 = 21 ].
"Fully specifying the table key components is required with TABLE KEY. So, the
"following statement is not possible.
"line = itab[ TABLE KEY sk comp2 = 20 ].
"------------------ KEY addition ------------------
"Using KEY and specifying all key components work like specifying TABLE KEY
line = itab[ KEY primary_key COMPONENTS comp1 = 1 ].
line = itab[ KEY primary_key comp1 = 1 ].
line = itab[ KEY sk COMPONENTS comp2 = 20 comp3 = 21 ].
line = itab[ KEY sk comp2 = 20 comp3 = 21 ].
"Unlike TABLE KEY, KEY does not enforce all key components to be specified
line = itab[ KEY sk comp2 = 20 ].
"In case of sorted and secondary table keys, other components not being part
"of the key can be specified
line = itab[ KEY primary_key comp1 = 1 comp4 = 'aaa' ].
line = itab[ KEY sk comp2 = 20 comp4 = 'bbb' ].
"The following statements are not possible. The initial, left part of
"the key must be specified. In the example case, it is comp2.
"line = itab[ KEY sk comp3 = 21 comp4 = 'bbb' ].
"line = itab[ KEY sk comp4 = 'bbb' ].
"The following statement triggers a syntax warning because the initial
"part of a table key is specified, but the key name is not specified.
"In this case, the search is not optimized as the component is not
"part of the primary table key of the sorted table. You may optimize
"it by specifying the key.
"line = itab[ comp2 = 10 ].
"The syntax warning can be suppressed by a pragma.
line = itab[ comp2 = 10 ] ##primkey[sk].
"Specifying the key name
line = itab[ KEY sk comp2 = 10 ].
"------------------ No TABLE KEY/KEY additions ------------------
"Specifying a free key search, but including all components of the primary
"table key
"For a sorted table as in the example, the search is fully optimized.
line = itab[ comp1 = 1 ].
"Partly optimized (only a part of the primary table key of the sorted
"example table is specified)
line = itab_so[ comp1 = 1 ]. |
Reading table lines using free keys |
"The search is and cannot be optimized as the component is not part of
"the primary table key of the sorted table. Plus, no appropriate
"secondary table key can be applied.
line = itab[ comp4 = 'ccc' ]. |
Using the read result in various positions |
This is to emphasize that table expressions can be used in various read positions. The snippet shows a few examples. Check the notes in the ABAP Keyword Documentation topics about the use. DATA(first_line) = itab[ 1 ].
"Using table expressions as arguments of built-in functions
IF line_exists( itab[ 1 ] ).
...
ELSE.
...
ENDIF.
ASSERT line_index( itab[ comp4 = 'ccc' ] ) = 3. |
Assigning table lines to a field symbol |
"Works like READ TABLE ... ASSIGNING ...
ASSIGN itab[ 2 ] TO FIELD-SYMBOL(<line>).
"Note: Table expressions do not set the sy-tabix
"value, except when used with ASSIGN.
ASSERT sy-tabix = 2.
"Note: Assigning a non-existent line results in sy-subrc = 4
"An exception is not raised.
ASSIGN itab[ 99 ] TO <line>.
ASSERT sy-subrc = 4. |
Data reference variables pointing to a table line |
DATA ref TYPE REF TO data.
ref = NEW s_demo( ).
ref->* = itab[ 1 ]. |
Specifying table expressions as operands in constructor expressions with VALUE and REF |
line = VALUE #( itab[ 2 ] ).
"Works like READ TABLE ... REFERENCE INTO ...
DATA(line_ref) = REF #( itab[ 3 ] ). |
Specifying a default value for lines that are not found to avoid an exception |
"Accessing a non-existent table line raises a catchable exception
TRY.
line = itab[ 4 ].
CATCH cx_sy_itab_line_not_found.
ENDTRY.
line = VALUE #( itab[ 4 ] OPTIONAL ).
line = VALUE #( itab[ 5 ] DEFAULT itab[ 1 ] ).
line = VALUE #( itab[ 6 ] DEFAULT VALUE #( ) ). |
Field symbols and dereferenced data references specified before the square brackets |
The previous examples use concrete internal table names specified before the square brackets. Field symbols and dereferenced data references are also possible. ASSIGN itab TO FIELD-SYMBOL(<tab>).
line = <tab>[ 1 ].
DATA dref TYPE REF TO ttyp.
dref = NEW #( ).
dref->* = itab.
line = dref->*[ 2 ]. |
Reading multiple lines |
Using table expressions, you can read entire lines individually. This is just to emphasize that when reading multiple lines sequentially, "Instead of something like this ...
DO.
ASSIGN itab[ sy-index ] TO FIELD-SYMBOL(<fs>).
IF sy-subrc = 0.
...
"<fs>-comp1 = ...
ELSE.
EXIT.
ENDIF.
ENDDO.
"... a LOOP statement, for example.
LOOP AT itab ASSIGNING <fs>.
...
"<fs>-comp1 = ...
ENDLOOP. |
Reading individual components of table lines |
"Reading individual components of table lines using chainings after the
"closing square bracket
"Read component via line read using ...
"... index
DATA(compa) = itab[ 1 ]-comp1.
"... table key
DATA(compb) = itab[ TABLE KEY primary_key comp1 = 1 ]-comp2.
DATA(compc) = itab[ TABLE KEY sk comp2 = 30 comp3 = 31 ]-comp1.
"... free key
DATA(compd) = itab[ comp4 = 'ccc' ]-comp1. |
Chaining table expressions in the context of nested internal tables |
"Creating deep internal table
TYPES: BEGIN OF s_sub,
comp1 TYPE i,
comp2 TYPE i,
END OF s_sub,
tab_type_sub TYPE TABLE OF s_sub WITH EMPTY KEY,
BEGIN OF s,
compa TYPE i,
compb TYPE TABLE OF tab_type_sub WITH EMPTY KEY,
END OF s,
tab_type TYPE TABLE OF s WITH EMPTY KEY.
"Expressions helpful when populating
DATA(deep_tab) = VALUE tab_type( ( compa = 1
compb = VALUE #( ( VALUE #( ( comp1 = 3 comp2 = 4 ) ( comp1 = 5 comp2 = 6 ) ) )
( VALUE #( ( comp1 = 7 comp2 = 8 ) ( comp1 = 9 comp2 = 10 ) ) ) ) )
( compa = 2
compb = VALUE #( ( VALUE #( ( comp1 = 11 comp2 = 12 ) ( comp1 = 13 comp2 = 14 ) ) )
( VALUE #( ( comp1 = 15 comp2 = 16 ) ( comp1 = 17 comp2 = 18 ) ) ) ) ) ).
DATA(num1) = deep_tab[ 2 ]-compb[ 1 ][ 2 ]-comp2.
ASSERT num1 = 14.
"Such a statement instead of, for example, multiple statements as follows.
READ TABLE deep_tab INDEX 2 INTO DATA(wa1).
READ TABLE wa1-compb INDEX 1 INTO DATA(wa2).
READ TABLE wa2 INTO DATA(wa3) INDEX 2.
DATA(num2) = wa3-comp2.
ASSERT num2 = num1. |
Table expression result having a reference type enabling chainings with the object component selector |
DATA itab_ref TYPE TABLE OF REF TO s_demo WITH EMPTY KEY.
itab_ref = VALUE #( ( NEW s_demo( comp1 = 1 comp2 = 30 comp3 = 31 comp4 = 'aaa' ) ) ).
"Reading entire line by dereferencing
DATA(deref_line) = itab_ref[ 1 ]->*.
"Reading component by dereferencing
DATA(dref_compa) = itab_ref[ 1 ]->comp3.
"The following syntax is also possible (dereferencing operator followed
"by the component selector).
DATA(dref_compb) = itab_ref[ 1 ]->*-comp4. |
Table expressions in write positions: Writes on the entire line |
Note that you cannot perform writes on entire lines in the context of key tables as key values cannot be changed. "The demo table is a key table. Therefore, writes on entire lines produce runtime errors.
"itab[ 3 ] = VALUE #( ).
"Creating a standard table
DATA itab_std TYPE TABLE OF s_demo WITH NON-UNIQUE KEY comp1 WITH NON-UNIQUE SORTED KEY sk COMPONENTS comp2 comp3.
itab_std = itab.
"Here, writes on entire lines are allowed.
itab_std[ 3 ] = VALUE #( comp1 = 123 comp4 = 'zzz' ).
CLEAR itab_std[ 3 ]. |
Table expressions in write positions: Writes on individual components |
itab[ 3 ]-comp4 = 'yyy'.
itab_ref[ 1 ]->comp3 = 123.
"No key value change allowed in key tables
"The following statement causes a runtime error.
"itab[ 1 ]-comp1 = 987.
"Key value change allowed for standard tables.
itab_std[ 3 ]-comp1 = 456. |
Dynamic key and component name specifications |
For more information about dynamic programming, refer to the Dynamic Programming cheat sheet. "Dynamic key and component specifications
line = itab[ KEY ('PRIMARY_KEY') COMPONENTS comp1 = 1 ].
line = itab[ KEY ('PRIMARY_KEY') comp1 = 1 ].
line = itab[ KEY ('SK') COMPONENTS comp2 = 20 comp3 = 21 ].
line = itab[ KEY ('SK') comp2 = 20 comp3 = 21 ].
line = itab[ KEY ('PRIMARY_KEY') ('COMP1') = 1 ].
line = itab[ KEY ('SK') ('COMP2') = 20 ('COMP3') = 21 ].
DATA(key_name) = 'PRIMARY_KEY'.
DATA(comp_name) = 'COMP1'.
line = itab[ KEY (key_name) (comp_name) = 1 ].
line = itab[ KEY primary_key (comp_name) = 1 ]. |
Pitfalls regarding table expressions |
🟢 Click to expand for example codeCLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
"-------------- Chaining table expressions -------------
"Creating a deep internal table
TYPES: BEGIN OF s_sub,
comp1 TYPE i,
comp2 TYPE i,
END OF s_sub,
tab_type_sub TYPE TABLE OF s_sub WITH EMPTY KEY,
BEGIN OF s,
compa TYPE i,
compb TYPE TABLE OF tab_type_sub WITH EMPTY KEY,
END OF s,
tab_type TYPE TABLE OF s WITH EMPTY KEY.
"Expressions helpful when populating
DATA(deep_tab) = VALUE tab_type( ( compa = 1
compb = VALUE #( ( VALUE #( ( comp1 = 3 comp2 = 4 ) ( comp1 = 5 comp2 = 6 ) ) )
( VALUE #( ( comp1 = 7 comp2 = 8 ) ( comp1 = 9 comp2 = 10 ) ) ) ) )
( compa = 2
compb = VALUE #( ( VALUE #( ( comp1 = 11 comp2 = 12 ) ( comp1 = 13 comp2 = 14 ) ) )
( VALUE #( ( comp1 = 15 comp2 = 16 ) ( comp1 = 17 comp2 = 18 ) ) ) ) ) ).
"Chained table expressions in the context of reading a value from a
"nested internal table
"Such a chaining works if the table expression result is a table itself.
"Such statements are fairly short, using few lines of code, however,
"they may be hard to understand and debug.
DATA(num1) = deep_tab[ 2 ]-compb[ 1 ][ 2 ]-comp2.
out->write( num1 ).
"A statement as above instead of multiple statements, for example, as follows.
READ TABLE deep_tab INDEX 2 INTO DATA(wa1).
READ TABLE wa1-compb INDEX 1 INTO DATA(wa2).
READ TABLE wa2 INTO DATA(wa3) INDEX 2.
DATA(num2) = wa3-comp2.
out->write( num2 ).
"-------------- Overusing table expression -------------
"The following examples performes many reads and writes
"using table expressions. The first example includes multiple
"table expressions, while the second one is differently designed,
"without many table expressions. The runtimes of the loops are
"stored, demonstrating that the second example has a significantly
"reduced runtime compared to the first example.
CONSTANTS: num_of_repetitions TYPE i VALUE 10,
num_of_assignment_runs TYPE i VALUE 1000.
TYPES: BEGIN OF struct,
a TYPE i,
b TYPE i,
c TYPE i,
d TYPE i,
e TYPE i,
f TYPE i,
END OF struct.
DATA it TYPE TABLE OF struct WITH EMPTY KEY.
it = VALUE #( ( a = 0 b = 0 c = 0 d = 0 e = 0 f = 0 )
( a = 1 b = 1 c = 1 d = 1 e = 1 f = 1 ) ).
DATA(it_original) = it.
DATA runtime_tab TYPE TABLE OF decfloat34 WITH EMPTY KEY.
DO num_of_repetitions TIMES.
DATA(ts1) = utclong_current( ).
DO num_of_assignment_runs TIMES.
it[ 1 ]-a = sy-index.
it[ 1 ]-b = sy-index.
it[ 1 ]-c = sy-index.
it[ 2 ]-d = sy-index.
it[ 2 ]-e = sy-index.
it[ 2 ]-f = sy-index.
it[ 1 ]-d = it[ 2 ]-d.
it[ 1 ]-e = it[ 2 ]-e.
it[ 1 ]-f = it[ 2 ]-f.
INSERT it[ 1 ] INTO TABLE it.
INSERT it[ 2 ] INTO TABLE it.
ENDDO.
DATA(ts2) = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = DATA(seconds) ).
APPEND seconds TO runtime_tab.
ENDDO.
SORT runtime_tab BY table_line ASCENDING.
out->write( `Multiple table expressions, fastest run:` ).
out->write( runtime_tab[ 1 ] ).
CLEAR runtime_tab.
it = it_original.
DO num_of_repetitions TIMES.
ts1 = utclong_current( ).
DO num_of_assignment_runs TIMES.
ASSIGN it[ 1 ] TO FIELD-SYMBOL(<fs1>).
ASSIGN it[ 2 ] TO FIELD-SYMBOL(<fs2>).
<fs1>-a = sy-index.
<fs1>-b = sy-index.
<fs1>-c = sy-index.
<fs2>-d = sy-index.
<fs2>-e = sy-index.
<fs2>-f = sy-index.
<fs1>-d = <fs2>-d.
<fs1>-e = <fs2>-e.
<fs1>-f = <fs2>-f.
INSERT <fs1> INTO TABLE it.
INSERT <fs2> INTO TABLE it.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
APPEND seconds TO runtime_tab.
ENDDO.
SORT runtime_tab BY table_line ASCENDING.
out->write( `Avoiding overusing table expressions, fastest run:` ).
out->write( runtime_tab[ 1 ] ).
ENDMETHOD.
ENDCLASS. |
This is relevant if you are not interested in the content of a table
line, but only want to find out whether a line exists that matches to the
index or key specifications. To do this, use a READ TABLE
statement with the TRANSPORTING NO FIELDS
addition. The
addition indicates that no actual content is to be read. If the search was
successful and an entry exists, the system field sy-subrc
is
set to 0.
A newer way to check the existence of a line is the predicate
function
line_exists( )
.
This function expects a table
expression as an argument.
See below for more on table expressions. Note that table expressions do not set system fields.
"Read using a key
READ TABLE it WITH KEY b = 2 TRANSPORTING NO FIELDS.
IF sy-subrc = 0.
...
ENDIF.
"Read using the index
READ TABLE it INDEX 1 TRANSPORTING NO FIELDS.
IF sy-subrc = 0.
...
ENDIF.
"Read using the key
IF line_exists( it[ b = 2 ] ).
...
ENDIF.
"Read using the index
IF line_exists( it[ 1 ] ).
...
ENDIF.
If you want to find out about the index of a line in an internal table, you can also make use of the READ TABLE
statement above. If
the line is found, the system field sy-tabix
is set to the number of the index. Otherwise, the built-in function
line_index( )
can be used. It returns the index of the found line or 0 if the line does not exist.
DATA(itab) = VALUE string_table( ( `aaa` ) ( `bbb` ) ).
READ TABLE itab WITH KEY table_line = `bbb` TRANSPORTING NO FIELDS.
"2
DATA(tabix) = sy-tabix.
"1
DATA(idx) = line_index( itab[ table_line = `aaa` ] ).
"Note: No primary table index with hashed tables
DATA(hashed_tab) = VALUE string_hashed_table( ( `a` ) ( `b` ) ( `c` ) ).
"-1
DATA(hashed_primary_idx) = line_index( hashed_tab[ table_line = `c` ] ).
"Index access in hashed tables only using a secondary table index
TYPES: BEGIN OF s,
comp1 TYPE i,
comp2 TYPE i,
END OF s,
ttype TYPE HASHED TABLE OF s WITH UNIQUE KEY comp1 WITH NON-UNIQUE SORTED KEY sk COMPONENTS comp2.
DATA(hashed_tab2) = VALUE ttype( ( comp1 = 1 comp2 = 10 )
( comp1 = 2 comp2 = 8 )
( comp1 = 3 comp2 = 9 ) ).
"3
DATA(hashed_secondary_idx) = line_index( hashed_tab2[ KEY sk comp2 = 10 ] ).
"1
hashed_secondary_idx = line_index( hashed_tab2[ KEY sk comp2 = 8 ] ).
lines( )
is another built-in function that you can use to check how many lines exist in an internal table. It returns an integer value.
DATA(itab) = VALUE string_table( ( `a` ) ( `b` ) ( `c` ) ( `d` ) ( `e` ) ).
"5
DATA(number_of_lines) = lines( itab ).
"Excursion: Finding out the number of lines in a table by specifying concrete
"component values, e.g. you want to find out how many lines exist in the table
"that have the value 1 for comp2
TYPES: BEGIN OF struct,
comp1 TYPE c LENGTH 3,
comp2 TYPE i,
END OF struct,
tab_type TYPE TABLE OF struct WITH EMPTY KEY.
DATA(it) = VALUE tab_type( ( comp1 = 'a' comp2 = 1 )
( comp1 = 'b' comp2 = 1 )
( comp1 = 'c' comp2 = 1 )
( comp1 = 'd' comp2 = 2 )
( comp1 = 'e' comp2 = 3 )
( comp1 = 'f' comp2 = 4 )
( comp1 = 'g' comp2 = 5 ) ).
"7
DATA(line_num) = lines( it ).
"Finding out the number of lines in a table by component value, e.g.
"using constructor expressions and specifying a WHERE clause.
"The example creates an new internal table inline using VALUE and a FOR loop,
"specified with a WHERE clause. The lines function is applied to the
"table created inline.
"3
DATA(line_num_filtered1) = lines( VALUE tab_type( FOR wa IN it WHERE ( comp2 = 1 ) ( wa ) ) ).
"Using the REDUCE operator
"The example adds 1 to the resulting integer if the comp2 value of the iterated line is greater than 1.
"The lines function is not relevant in the example.
"4
DATA(line_num_filtered2) = REDUCE i( INIT var = 0
FOR <line> IN it
WHERE ( comp2 > 1 )
NEXT var += 1 ).
"Using the FILTER operator
"Note: The source table must have at least one sorted key or a hash key for accessing.
"If the table does not have such a primary table key, a secondary table key must be available.
TYPES: tab_type_sorted TYPE TABLE OF struct with NON-UNIQUE SORTED KEY sec_key COMPONENTS comp2.
DATA it_sorted type tab_type_sorted.
it_sorted = it.
"The example creates an new internal table inline using FILTER,
"specified with a WHERE clause. The lines function is applied to the
"table created inline.
"3
DATA(line_num_filtered3) = lines( FILTER #( it_sorted USING KEY sec_key WHERE comp2 = 1 ) ).
"4
DATA(line_num_filtered4) = lines( FILTER #( it_sorted USING KEY sec_key WHERE comp2 > 1 ) ).
Using Runtime Type Identification (RTTI), you can get type information on internal tables and table types at runtime
For more information, see the Dynamic Programming ABAP cheat sheet.
RTTI example:
TYPES tab_type TYPE SORTED TABLE OF zdemo_abap_flsch
WITH UNIQUE KEY carrid connid
WITH NON-UNIQUE SORTED KEY sec_key ALIAS sk COMPONENTS countryfr cityto.
DATA itab TYPE tab_type.
DATA(tdo_d) = cl_abap_typedescr=>describe_by_data( itab ).
"DATA(tdo_d) = cl_abap_typedescr=>describe_by_name( 'TAB_TYPE' ).
"Cast to get more specific information
DATA(tdo_itab) = CAST cl_abap_tabledescr( cl_abap_typedescr=>describe_by_data( itab ) ).
"DATA(tdo_itab) = CAST cl_abap_tabledescr( tdo_d ).
DATA(type_category_itab) = tdo_itab->kind.
DATA(relative_name_itab) = tdo_itab->get_relative_name( ).
... "Explore more options by positioning the cursor behind -> and choosing CTRL + Space
DATA(table_kind_itab) = tdo_itab->table_kind.
DATA(table_keys_itab) = tdo_itab->key.
DATA(table_keys_more_details_itab) = tdo_itab->get_keys( ).
DATA(table_has_unique_key_itab) = tdo_itab->has_unique_key.
DATA(table_key_alias_itab) = tdo_itab->get_key_aliases( ).
DATA(line_type_itab) = tdo_itab->get_table_line_type( ).
DATA(table_component_info_itab) = CAST cl_abap_structdescr( tdo_itab->get_table_line_type( ) ).
DATA(table_components_itab) = CAST cl_abap_structdescr( tdo_itab->get_table_line_type( ) )->components.
DATA(table_comps_more_info_itab) = CAST cl_abap_structdescr( tdo_itab->get_table_line_type( ) )->get_components( ).
DATA(applies_to_data_itab) = tdo_itab->applies_to_data( VALUE tab_type( ) ).
If you are interested not only in single table lines, but in the entire
table content or in specific parts of it, you can use LOOP AT
statements to process table lines sequentially. As above, you
can use multiple options for target areas: work area, field
symbol, data reference. There are multiple additions to the LOOP AT
statements to further restrict the table content to be processed.
Simple form:
"The target is an existing work area.
DATA wa LIKE LINE OF it.
LOOP AT it INTO wa.
"No addition of the loop statement; all lines are processed
"Statements in this block are relevant for each individual table line.
...
ENDLOOP.
"Work area declared inline
LOOP AT itab INTO DATA(wa_inl).
...
ENDLOOP.
"Field symbols
FIELD-SYMBOLS <fs> LIKE LINE OF it.
LOOP AT it ASSIGNING <fs>.
...
ENDLOOP.
LOOP AT it ASSIGNING FIELD-SYMBOL(<fs_inl>).
...
ENDLOOP.
"Data reference variables
DATA dref TYPE REF TO dbtab.
LOOP AT it REFERENCE INTO dref.
...
ENDLOOP.
LOOP AT it REFERENCE INTO DATA(dref_inl).
...
ENDLOOP.
"Table key specification (snippet uses example table from above)
"The specified table key affects the order in which the table lines
"are accessed and the evaluation of the other conditions.
LOOP AT it INTO wa USING KEY primary_key.
"LOOP AT it INTO wa USING KEY pk. "primary key alias
"LOOP AT it INTO wa USING KEY sec_key. "secondary key
"LOOP AT it INTO wa USING KEY sk. "secondary key alias
...
ENDLOOP.
- The order in which tables are iterated depends on the table category.
- Note the
STEP
addition, which is also available for other ABAP statements.
- Note the
- Index tables are looped over in ascending order by the index.
- Hashed tables are looped in the order in which the lines were added to the table. You can also sort the table before the loop.
- During the loop, the system field
sy-tabix
is set to the number of the currently processed table line. This is not true for hashed tables. There,sy-tabix
is0
. - Note that if you want to work with the value of
sy-tabix
, you should do so immediately after theLOOP
statement to avoid possible overwriting in statements contained in the loop block.
The additions of LOOP
statements come into play when you want to restrict the table content to be respected by the loop because
you do not want to loop over the entire table. Note that the examples only show work areas as target objects to hold the table line read.
Other options are possible as shown above.
"FROM/TO: Only for index tables
"Specifying an index range
LOOP AT it INTO wa FROM 2 TO 5.
...
ENDLOOP.
"From specified line until the end
LOOP AT it INTO wa FROM 2.
...
ENDLOOP.
"From first line until the specified line
LOOP AT it INTO wa TO 5.
...
ENDLOOP.
"WHERE clause: Restricting lines based on logical expression
LOOP AT it INTO wa WHERE a > 1 AND b < 4.
...
ENDLOOP.
"No interest in the table content; only relevant system fields are populated
"Mandatory WHERE clause
LOOP AT it TRANSPORTING NO FIELDS WHERE a < 5.
...
ENDLOOP.
Find more information in the STEP
topic. The addition is also available for other ABAP statements.
"STEP addition for defining the step size and the direction of the loop
"- Step size: Specified by the absolute value of an integer
"- Direction: Specified by a positive (forward loop) or negative
" (loop in reverse order) integer
"Reversing the loop order using a negative integer
"Each line is read indicated by the absolute value 1
LOOP AT it INTO wa STEP -1.
...
ENDLOOP.
"Forward loop by specifiying a positive integer
"In the example, every second line is read.
"Note: Omitting STEP means STEP 1 by default.
LOOP AT it INTO wa STEP 2.
...
ENDLOOP.
"STEP with other additions
"The example uses the additions FROM and TO.
"Note: If the value after STEP is negative, the value
"after FROM must be greater than the value after TO.
LOOP AT it INTO wa FROM 6 TO 3 STEP -2.
...
ENDLOOP.
Iteration expressions with FOR
as part of certain constructor expressions allow you to create content of an internal table by evaluating one or more source tables.
TYPES ty_int_tab TYPE TABLE OF i WITH EMPTY KEY.
DATA(int_table_a) = VALUE ty_int_tab( ( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ).
DATA int_table_b TYPE ty_int_tab.
int_table_b = VALUE #( FOR wa_b IN int_table_a ( wa_b * 2 ) ).
"Table Content: 2 / 4 / 6 / 8 / 10
"Instead of, for example, a LOOP statement as follows:
DATA int_table_c TYPE ty_int_tab.
LOOP AT int_table_a INTO DATA(wa_c).
INSERT wa_c * 3 INTO TABLE int_table_c.
ENDLOOP.
"Table content: 3 / 6 / 9 / 12 / 15
The expressions are covered in the cheat sheet Constructor Expressions:
- Iteration Expressions Using FOR
- Special reduction operator
REDUCE
that is based on iteration expressions: REDUCE
ABAP keywords such as CONTINUE
, CHECK
, and EXIT
, are available to exit and interrupt loops. Find more information in the Program Flow Logic cheat sheet.
In the following example, the loop is exited using the EXIT
statement when a certain condition is met.
DATA(str_table) = VALUE string_table( ( `a` ) ( `b` ) ( `c` ) ( `d` ) ( `e` ) ( `f` ) ).
LOOP AT str_table INTO DATA(wa).
DATA(tabix) = sy-tabix.
IF wa = `e`.
EXIT.
ENDIF.
ENDLOOP.
ASSERT tabix = 5.
- In a loop, you can change the values of the current table line's components. You can also delete or insert entire lines.
- The position of these new or deleted lines is determined by the table index or a sorted key. For hashed tables or hash keys, the position is determined by the insertion order (though this can be changed with the
SORT
statement). - After the currently processed table line, ...
- inserting a line: The new line is processed in the next loop pass (watch out for a non-terminating loop).
- deleting a line: The deleted line is not processed in the next loop pass.
- Before the currently processed table line, ...
- inserting a line: The internal loop counter (
sy-tabix
) increases accordingly (in the case of index tables, sorted keys). - deleting a line: The internal loop counter (
sy-tabix
) decreases accordingly.
- inserting a line: The internal loop counter (
- You cannot replace or clear the entire table body in a loop. If detected statically, a syntax error is displayed.
The following examples explore the insertion and deletion of lines when looping across internal tables:
"Creating and populating a demo standard internal table to
"work with in the example loops
TYPES: BEGIN OF s_loop_mod,
text TYPE string,
num TYPE i,
END OF s_loop_mod,
t_loop_mod TYPE TABLE OF s_loop_mod WITH EMPTY KEY.
"Inserting 10 entries into the demo table
DATA(itab_original) = VALUE t_loop_mod( FOR x = 1 WHILE x <= 10 ( text = x ) ).
DATA(itab) = itab_original.
"---------- Inserting a line after the current line ----------
"The example inserts a line after the currently processed line
"using an INSERT statement and specifying the index value
"(sy-tabix value + 1). The 'num' component is assigned the
"current sy-tabix value.
"Note: In all statements, the sy-tabix value is stored in a
"variable right after the LOOP statement. Assume that multiple
"statements are included before the statement that actually uses
"the current sy-tabix value. Other statements that potentially
"change the sy-tabix value might interfere.
"An EXIT statement takes care of exiting the loop. In the example,
"all values of the 'num' component in the original table lines
"(except the first line) are initial as the loop is exited.
LOOP AT itab ASSIGNING FIELD-SYMBOL(<fs>).
DATA(tabix) = sy-tabix.
<fs>-num = tabix.
INSERT VALUE #( text = tabix ) INTO itab INDEX tabix + 1.
IF tabix = 50.
EXIT.
ENDIF.
ENDLOOP.
*Internal table content (parts of it):
*TEXT NUM
*1 1
*1 2
*2 3
*3 4
*4 5
*...
*48 49
*49 50
*50 0
*2 0
*3 0
*4 0
*...
"---------- Deleting a line after the current line ----------
"The example deletes a line after the current line using a
"DELETE statement and the INDEX addition. The index value
"is specified using the current sy-tabix value + 1.
"The 'num' value in the resulting internal table includes
"the sy-tabix value.
itab = itab_original.
LOOP AT itab ASSIGNING <fs>.
tabix = sy-tabix.
<fs>-num = tabix.
DELETE itab INDEX tabix + 1.
ENDLOOP.
*Internal table content:
*TEXT NUM
*1 1
*3 2
*5 3
*7 4
*9 5
"---------- Inserting a line before the current line ----------
"The example insert a line before the currently processed line using
"an INSERT statement. The current sy-tabix value is used as INDEX value,
"moving down the currently processed table line one position.
"In that case, the sy-tabix value increases accordingly.
"Logic:
"- For example, the first line is processed, sy-tabix has the value 1.
"- A line is inserted at this position, moving down the currently processed line
" one position. The moved line is then in the second position (as a new line exists
" in the first position).
"- In the next loop pass, the loop is continued with the third line, i.e.
" sy-tabix has the value 3 in the second loop pass.
"The example includes modifications of the table components. The 'num' value
"is assigned the table index value after inserting the new line. The
"'num' value of existing table lines includes the value of the index before
"inserting the new line + 1.
itab = itab_original.
FIELD-SYMBOLS <line> TYPE s_loop_mod.
DATA new_line_counter TYPE i.
DATA tabix_copy TYPE i.
LOOP AT itab ASSIGNING <fs>.
tabix = sy-tabix.
new_line_counter += 1.
"Asserting that sy-tabix value has changed accordingly.
IF tabix <> 1.
ASSERT tabix = tabix_copy + 2.
ENDIF.
DATA(new_line_text) = |---- New line { new_line_counter } ----|.
INSERT VALUE #( text = new_line_text ) INTO itab INDEX tabix ASSIGNING <line>.
DATA(idx_new) = line_index( itab[ text = new_line_text num = 0 ] ).
<line>-num = idx_new.
DATA(idx_existing) = line_index( itab[ text = <fs>-text num = 0 ] ).
DATA(new_text) = |{ <fs>-text }(existing line, index before insertion: { tabix })|.
<fs>-text = new_text.
<fs>-num = idx_existing.
tabix_copy = tabix.
ENDLOOP.
*Internal table content:
*TEXT NUM
*---- New line 1 ---- 1
*1 (existing line, index before insertion: 1) 2
*---- New line 2 ---- 3
*2 (existing line, index before insertion: 3) 4
*---- New line 3 ---- 5
*3 (existing line, index before insertion: 5) 6
*---- New line 4 ---- 7
*4 (existing line, index before insertion: 7) 8
*---- New line 5 ---- 9
*5 (existing line, index before insertion: 9) 10
*---- New line 6 ---- 11
*6 (existing line, index before insertion: 11) 12
*---- New line 7 ---- 13
*7 (existing line, index before insertion: 13) 14
*---- New line 8 ---- 15
*8 (existing line, index before insertion: 15) 16
*---- New line 9 ---- 17
*9 (existing line, index before insertion: 17) 18
*---- New line 10 ---- 19
*10 (existing line, index before insertion: 19) 20
"---------- Deleting a line before the current line ----------
"The example explores the deletion of a line before the currently
"processed line. The previous line in the table is deleted if
"the value of 'text' (an integer was inserted) is an even number.
"The DELETE statement specifies the index with the current sy-tabix
"value - 1. On deletion, the sy-tabix value is decreased accordingly.
"Before a potential deletion, the currently processed table line is
"copied to another table to visualize the current sy-tabix value in
"the 'num' component.
itab = itab_original.
DATA itab_copy LIKE itab.
LOOP AT itab ASSIGNING <fs>.
tabix = sy-tabix.
<fs>-num = tabix.
INSERT <fs> INTO TABLE itab_copy.
TRY.
IF CONV i( <fs>-text ) MOD 2 = 0.
DELETE itab INDEX tabix - 1.
ENDIF.
CATCH cx_sy_conversion_no_number.
ENDTRY.
ENDLOOP.
*Internal table content (itab):
*TEXT NUM
*2 2
*4 3
*6 4
*8 5
*10 6
*Internal table content (itab_copy):
*TEXT NUM
*1 1
*2 2
*3 2
*4 3
*5 3
*6 4
*7 4
*8 5
*9 5
*10 6
"---------- Deleting the currently processed table line ----------
"The example explores deleting the currently processed table line using
"a string table. So, the DELETE statement specifies the current sy-tabix
"value for INDEX. In that case, the next line moves up one position, and
"the sy-tabix value remains the same, i.e. when sy-tabix is 2, and the
"line is deleted, the value remains 2 and processes the line moved up.
"Creating and populating a demo internal table
DATA(str_tab) = VALUE string_table( ( `a` ) ( `#` ) ( `c` ) ( `#` ) ( `e` )
( `f` ) ( `g` ) ( `#` ) ( `i` ) ( `j` ) ).
LOOP AT str_tab REFERENCE INTO DATA(dref).
tabix = sy-tabix.
IF dref->* CS `#`.
DELETE str_tab INDEX tabix.
ENDIF.
ENDLOOP.
*Internal table content:
*a
*c
*e
*f
*g
*i
*j
"---------- Statements clearing the entire internal table are not allowed in loops ----------
"The entire internal table cannot be deleted within loops.
"The following statements commented out are not possible.
LOOP AT str_tab REFERENCE INTO dref.
"CLEAR str_tab.
"str_tab = VALUE #( ).
ENDLOOP.
Adding multiple lines from a database table to an internal table using
SELECT
,
for example, based on a condition. In the case below, the internal table
is created inline.
SELECT FROM dbtab
FIELDS comp1, comp2 ...
WHERE ...
INTO TABLE @DATA(itab_sel).
Adding multiple lines from a database table using SELECT
, for example, based on a condition when the database table has a line type that is incompatible with the internal table. The *
character means that all fields are selected. The other examples define specific fields.
The APPENDING CORRESPONDING FIELDS INTO TABLE
addition appends the selected data to the end of the table without deleting existing
table entries. The INTO CORRESPONDING FIELDS OF TABLE
addition adds lines and deletes existing table entries.
SELECT FROM dbtab2
FIELDS *
WHERE ...
APPENDING CORRESPONDING FIELDS OF TABLE @itab.
SELECT FROM dbtab2
FIELDS *
WHERE ...
INTO CORRESPONDING FIELDS OF TABLE @itab.
Combining data from multiple database tables into one internal table using an inner
join.
The following example uses the INNER JOIN
addition. Note that the field list includes fields from both tables. The fields are referred to using ~
.
SELECT db1~comp1, db1~comp2, db2~comp_abc, db2~comp_xyz ...
FROM db1
INNER JOIN db2 ON db1~comp1 = db2~comp1
INTO TABLE @DATA(it_join_result).
Populating an internal table from a database table using
subqueries.
The following two examples populate an internal table from a database table. In the first example, a subquery is specified in the
WHERE
clause with the NOT IN
addition. It checks whether a value matches a value in a set of values
specified in parentheses. The second example populates an internal table depending on data in another table. A subquery with the EXISTS
addition is specified in
the WHERE
clause. In this
case, the result of the subquery, which is another
SELECT
statement, is checked to see if an entry exists in
a table based on the specified conditions.
SELECT comp1, comp2, ...
FROM dbtab
WHERE comp1 NOT IN ( a, b, c ... )
INTO TABLE @DATA(it_subquery_result1).
SELECT comp1, comp2, ...
FROM db1
WHERE EXISTS ( SELECT 'X' FROM db2
WHERE comp1 = db1~comp1 )
INTO TABLE @DATA(it_subquery_result2).
Populating an internal table from a table based on the existence of data in
another table using the FOR ALL ENTRIES
addition.
💡 Note
Make sure that the internal table you are reading from is not initial. Therefore, it is recommended that you use a subquery as shown above:... ( SELECT ... FROM ... WHERE ... ) ...
.
IF itab IS NOT INITIAL.
SELECT dbtab~comp1, dbtab~comp2, ...
FROM dbtab
FOR ALL ENTRIES IN @itab
WHERE comp1 = @itab-comp1
INTO TABLE @DATA(it_select_result).
ENDIF.
-
You can use internal tables as data sources in
SELECT
statements.SELECT comp1, comp2, ... FROM @itab AS it_alias WHERE ... INTO TABLE @DATA(itab_sel).
-
Internal tables are specified as host variables prefixed by
@
and provided with an alias name. -
Deep and nested structured types are not allowed. Structured types cannot include strings, reference types, or internal tables. Exception: The type
string
is allowed if it is declared using a reference to the built-in dictionary typesstring
. -
When used as data sources in
SELECT
statements, internal tables are treated like DDIC database tables.- This is the case even if they are not passed to the database.
- They are considered as client-independent tables, and the first column is not considered as a client column.
- The ABAP types of the columns are mapped to appropriate DDIC built-in types.
-
Using
SELECT
statements with internal tables has significant advantages:- You can leverage the extensive functionalities that ABAP SQL provides with
SELECT
statements, such as aggregate expressions. - They can serve as alternatives to
READ TABLE
orLOOP AT
statements, provided the data can be processed on the AS ABAP by the ABAP SQL engine and the SQL functionality surpasses the standard functionality of these ABAP statements.
- You can leverage the extensive functionalities that ABAP SQL provides with
Notes and restrictions:
- In ABAP, database data is buffered in a table buffer (internally, this happens in internal tables in the shared memory of the ABAP server).
- During read access, it is checked if the data is in the buffer, and if so, a read happens directly from there. If not, the data is first loaded into the buffer.
- The ABAP SQL engine is involved in the read process. It processes reads and is used when tabular data is read (with a
SELECT
statement). This includes both buffered data from database tables in the table buffer and also internal tables of the current internal session. - Which means the ABAP SQL engine processes queries located on the AS ABAP, ABAP SQL is executed in the said buffer on the AS ABAP, not directly on the database.
- However, if a
SELECT
statement includes elements the ABAP SQL engine cannot handle in case of internal tables, the internal table data transfers to a temporary database table for query execution. - Yet, only the data of one internal table can be transferred to the database. Thus, if a query involves multiple internal tables, it can only be executed if the ABAP SQL engine can manage it. This means that if data from more than one internal table must be transferred to the database, the query will not function. Similarly, joins of database tables and internal tables can only specify one internal table whose data can be passed to the database.
- If the compiler detects a statement the ABAP SQL engine cannot process, a syntax warning appears. To suppress this warning, use the pragma
##itab_db_select
. - Find more information on ...
- the restrictions here.
- the various ABAP SQL functionalities in the ABAP Keyword Documentation and in the ABAP SQL cheat sheet. The following code snippets cover a selection.
The following example explores various SELECT
queries with internal tables as data sources. To try it out, create a demo class named zcl_some_class
and paste the code into it. After activation, choose F9 in ADT to execute the class. The example uses objects of the ABAP cheat sheets repository and is set up to display output in the console.
Example:
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
"----------- Exploiting ABAP SQL functionality with internal tables -----------
TYPES int_tab_type TYPE TABLE OF i WITH EMPTY KEY.
DATA(itab_a) = VALUE int_tab_type( ( 1 ) ( 32 ) ( 100 ) ( -24 ) ( 17 ) ( 99 ) ).
"SELECT query with an internal table as data source
"The example uses an aggregate expression. It is statically
"detected that the query cannot be processed by the ABAP SQL
"engine. The data must be passed to the database. Consequently,
"a syntax warning is displayed. It can be suppressed by a pragma.
SELECT MAX( table_line ) AS max_val
FROM @itab_a AS it
INTO @DATA(max_a).
"100
SELECT MAX( table_line ) AS max_val ##itab_db_select
FROM @itab_a AS it
INTO @DATA(max_b).
out->write( max_a ).
out->write( max_b ).
"Using the LIKE addition in the WHERE clause to extract internal table
"entries matching a specific pattern.
TYPES: BEGIN OF s1,
a TYPE c LENGTH 3,
b TYPE i,
END OF s1,
it_type_1 TYPE TABLE OF s1 WITH EMPTY KEY.
DATA(itab_b) = VALUE it_type_1( ( a = 'abc' b = 1 )
( a = 'zbc' b = 2 )
( a = 'bde' b = 3 )
( a = 'yde' b = 4 ) ).
SELECT a, b
FROM @itab_b AS it_alias
WHERE a LIKE '%bc'
INTO TABLE @DATA(select_like_result).
*A B
*abc 1
*zbc 2
out->write( select_like_result ).
"----------- Using a SELECT loop with an internal table as data source -----------
TYPES: BEGIN OF s2,
comp1 TYPE c LENGTH 2,
comp2 TYPE i,
END OF s2,
it_type_2 TYPE TABLE OF s2 WITH EMPTY KEY.
DATA(itab_c) = VALUE it_type_2( ( comp1 = 'aa' comp2 = 2 )
( comp1 = 'zz' comp2 = 9 )
( comp1 = 'dd' comp2 = 1 )
( comp1 = 'rr' comp2 = 7 )
( comp1 = 'tt' comp2 = 5 )
( comp1 = 'bb' comp2 = 6 ) ).
DATA itab_d TYPE int_tab_type.
"The following SELECT loop specifies an internal table as data source.
"The loop sequence is defined by a sort order. Such a functionality is
"not available with LOOP AT.
SELECT comp2
FROM @itab_c AS it
ORDER BY comp2 DESCENDING
INTO @DATA(wa).
INSERT wa INTO TABLE itab_d.
ENDSELECT.
*9
*7
*6
*5
*2
*1
out->write( itab_d ).
"------------------- Joins with internal tables -------------------
TYPES: BEGIN OF s3,
a TYPE c LENGTH 3,
b TYPE c LENGTH 3,
c TYPE i,
END OF s3,
it_type_3 TYPE TABLE OF s3 WITH EMPTY KEY.
DATA(itab_e) = VALUE it_type_3( ( a = 'aaa' b = 'bbb' c = 1 )
( a = 'ccc' b = 'ddd' c = 1 )
( a = 'eee' b = 'fff' c = 2 ) ).
DATA(itab_f) = VALUE it_type_3( ( a = 'ggg' b = 'hhh' c = 1 )
( a = 'iii' b = 'jjj' c = 1 )
( a = 'kkk' b = 'lll' c = 3 ) ).
"No syntax warning. The internal tables can be processed by the
"ABAP SQL engine.
SELECT it_alias1~a, it_alias2~b
FROM @itab_e AS it_alias1
INNER JOIN @itab_f AS it_alias2 ON it_alias1~c = it_alias2~c
INTO TABLE @DATA(itab_g).
*A B
*aaa hhh
*aaa jjj
*ccc hhh
*ccc jjj
out->write( itab_g ).
"Join with a database table and an internal table
"Preparing a demo database table and an internal table
DELETE FROM zdemo_abap_tab1.
INSERT zdemo_abap_tab1 FROM TABLE @( VALUE #( ( key_field = 1 char1 = 'aaa' )
( key_field = 2 char1 = 'bbb' )
( key_field = 3 char1 = 'ccc' ) ) ).
TYPES it_type_4 TYPE TABLE OF zdemo_abap_tab1 WITH EMPTY KEY.
DATA(itab_h) = VALUE it_type_4( ( key_field = 1 char2 = 'zzz' )
( key_field = 2 char2 = 'yyy' ) ).
SELECT db~key_field, db~char1, it~char2
FROM zdemo_abap_tab1 AS db
INNER JOIN @itab_h AS it ON it~key_field = db~key_field
INTO TABLE @DATA(itab_i).
*KEY_FIELD CHAR1 CHAR2
*1 aaa zzz
*2 bbb yyy
out->write( itab_i ).
ENDMETHOD.
ENDCLASS.
- This excursion is intended to underscore the restrictions mentioned above and in the documentation in more detail when selecting from internal tables.
- Components having deep types, such as strings, cannot be included, for example, in the
SELECT
list orWHERE
clause. - Note that only those fields are checked (and sent to the database) that are actually used (as shown in the example below).
- However, the type string is allowed if it is declared using the built-in dictionary type
sstring
(for example, a component typed with a data element or a CDS simple type that usessstring
).
The following example demonstrates various SELECT
statements. A demo internal table has a component that is typed with a CDS simple type, which can be created as follows:
- In ADT, right-click your pacakage, and choose New -> Other Repository Object
- Insert type and select Type under Core Data Services.
- Choose Next and provide a name (e.g.
zdemo_abap_string
) and a description. - Choose Finish.
- Find more information on CDS simple types here.
The code of the CDS simple type may look as follows:
@EndUserText.label: 'String type'
define type zdemo_abap_string: abap.sstring( 1333 );
Commented code example:
"------ Internal table having a component of type string ------
"Creating and populating an internal table that includes a component that is
"typed with type string
TYPES: BEGIN OF s1,
comp1 TYPE i,
comp2 TYPE c LENGTH 3,
comp3 TYPE string,
END OF s1,
tab_type_1 TYPE TABLE OF s1 WITH EMPTY KEY.
DATA(itab1) = VALUE tab_type_1( ( comp1 = 1 comp2 = 'aaa' comp3 = `ABAP` ) ).
"Columns of type string cannot be used in SELECT statements that
"select from internal tables (* selects all fields). Therefore, the
"following statements are commented out.
"SELECT SINGLE * FROM @itab1 AS it WHERE comp1 = 1 INTO @DATA(res1).
"SELECT SINGLE comp1 FROM @itab1 AS it where comp3 = `ABAP` INTO @DATA(res2).
"SELECT SINGLE comp3 FROM @itab1 AS it where comp1 = 1 INTO @DATA(res3).
"However, the following statements work. No component of type string is involved.
SELECT SINGLE comp1 FROM @itab1 AS it WHERE comp2 = 'aaa' INTO @DATA(res4).
SELECT SINGLE comp1, comp2 FROM @itab1 AS it INTO @DATA(res5).
"--- Internal table having a component typed with a CDS simple type (sstring) ---
"Creating and populating an internal table that includes a component that is typed
"with a CDS simple type (sstring). Note: Built-in DDIC types such as sstring cannot
"directly be used in ABAP statements, except for typed literals.
TYPES: BEGIN OF s2,
comp1 TYPE i,
comp2 TYPE c LENGTH 3,
comp3 TYPE zdemo_abap_string,
END OF s2,
tab_type_2 TYPE TABLE OF s2 WITH EMPTY KEY.
DATA(itab2) = VALUE tab_type_2( ( comp1 = 1 comp2 = 'aaa' comp3 = `ABAP` ) ).
"Unlike above, the following SELECT statements are possible
SELECT SINGLE * FROM @itab2 AS it WHERE comp1 = 1 INTO @DATA(res6).
SELECT SINGLE comp1 FROM @itab2 AS it WHERE comp3 = `ABAP` INTO @DATA(res7).
SELECT SINGLE comp1, comp3 FROM @itab2 AS it WHERE comp2 = 'aaa' AND comp3 = `ABAP` INTO @DATA(res8).
"Note: `ABAP` represents a literal of type string. When specifying it here
"like it is specified above, there is an implicit conversion. The following
"example uses a typed literal.
SELECT SINGLE comp1, comp2, comp3 FROM @itab2 AS it WHERE comp3 = sstring`ABAP` INTO @DATA(res9).
The following code snippets demonstrate joining/merging the content of two simple internal tables into another table. There may be several ways to achieve this. Here, the intention is to give an idea and, in particular, to emphasize SQL functionalities also available for internal tables (note the restrictions mentioned above and in the documentation).
Assumptions:
- The target table is either created inline or exists and includes components from the source tables or components that can be mapped.
- One or more components are common between the source tables to perform a join or merge the table content.
The code snippet shows a selection of syntax options and includes the following statements:
- Joins with
SELECT
(INNER JOIN
,LEFT OUTER JOIN
, CTE) - Loops with
LOOP AT
statements,FOR
loops using theVALUE
andREDUCE
operators
"Creating two internal tables whose content will be joined. The shared
"value is represented by the key1 and key2 components.
"Sorted tables are used in the example (having key1/key2 as unique keys)
"to have unique values to perform joins.
TYPES: BEGIN OF s1,
key1 TYPE i,
a TYPE c LENGTH 1,
b TYPE c LENGTH 1,
c TYPE c LENGTH 1,
END OF s1,
tab_type1 TYPE SORTED TABLE OF s1 WITH UNIQUE KEY key1,
BEGIN OF s2,
key2 TYPE i,
d TYPE c LENGTH 1,
e TYPE c LENGTH 1,
END OF s2,
tab_type2 TYPE SORTED TABLE OF s2 WITH UNIQUE KEY key2.
"Populating demo internal tables
DATA(itab1) = VALUE tab_type1( ( key1 = 1 a = 'a' b = 'b' c = 'c' )
( key1 = 2 a = 'd' b = 'e' c = 'f' )
( key1 = 3 a = 'g' b = 'h' c = 'i' ) ).
DATA(itab2) = VALUE tab_type2( ( key2 = 1 d = `j` e = `k` )
( key2 = 2 d = `l` e = `m` ) ).
"SELECT statement, inner join
"Note: With the inner join, the target table contains all
"combinations of rows for whose columns the join condition
"is true.
SELECT a~key1, a~a, a~b, b~d, b~e
FROM @itab1 AS a
INNER JOIN @itab2 AS b ON a~key1 = b~key2
INTO TABLE @DATA(itab3).
*Result
*KEY1 A B D E
*1 a b j k
*2 d e l m
"SELECT statement, left outer join
"In contrast to the inner join above, the target table here
"also contains the table row of the first table for which
"no equivalent row exists in the second table.
SELECT a~key1, a~a, a~b, b~d, b~e
FROM @itab1 AS a
LEFT OUTER JOIN @itab2 AS b ON a~key1 = b~key2
INTO TABLE @DATA(itab4).
*Result
*KEY1 A B D E
*1 a b j k
*2 d e l m
*3 g h
"------------------------ NOTE -----------------------------------
"--- The following statements produce the same result as the -----
"--- previous example (itab4). -----------------------------------
"-----------------------------------------------------------------
"Common table expression
WITH +it1 AS ( SELECT a~key1, a~a, a~b FROM @itab1 AS a ),
+it2 AS ( SELECT b~key2, b~d, b~e FROM @itab2 AS b )
SELECT +it1~key1, +it1~a, +it1~b, +it2~d, +it2~e FROM +it1 LEFT JOIN +it2 ON +it1~key1 = +it2~key2
INTO TABLE @DATA(itab5).
"LOOP statements
"Using the CORRESPONDING operator to assign identically named components,
"BASE retains existing content
"The assignment with CORRESPONDING ... BASE ... includes a table expression
"in which table lines are read and inserted based on the key mapping. With the
"OPTIONAL addition, errors can be avoided if a line does not exist.
DATA itab6 LIKE itab4.
LOOP AT itab1 INTO DATA(wa1).
INSERT CORRESPONDING #( wa1 ) INTO TABLE itab6 REFERENCE INTO DATA(ref).
ref->* = CORRESPONDING #( BASE ( ref->* ) VALUE #( itab2[ key2 = ref->key1 ] OPTIONAL ) ).
ENDLOOP.
"Assume the second table's shared component was also key1. In the second CORRESPONDING
"you could then work with the EXCEPT addition to not overwrite the identicall named
"component.
"Example similar to the previous one
"Also here, a table expression is used to read a line from
"the second internal table. The INSERT statement (without
"CORRESPONDING) includes the concrete value assignments
"with the VALUE operator.
DATA itab7 LIKE itab4.
LOOP AT itab1 INTO DATA(wa2).
DATA(line) = VALUE #( itab2[ key2 = wa2-key1 ] OPTIONAL ).
INSERT VALUE #( key1 = wa2-key1
a = wa2-a
b = wa2-b
d = line-d
e = line-e ) INTO TABLE itab7.
ENDLOOP.
"Example using a FOR loop with the VALUE operator
TYPES tt_type3 LIKE itab4.
DATA(itab8) = VALUE tt_type3( FOR wa3 IN itab1
( key1 = wa3-key1
a = wa3-a
b = wa3-b
d = VALUE #( itab2[ key2 = wa3-key1 ]-d OPTIONAL )
e = VALUE #( itab2[ key2 = wa3-key1 ]-e OPTIONAL ) ) ).
"Similar example that includes a LET expression
DATA(itab9) = VALUE tt_type3( FOR wa4 IN itab1
LET tab_line = VALUE #( itab2[ key2 = wa4-key1 ] OPTIONAL ) IN
( key1 = wa4-key1
a = wa4-a
b = wa4-b
d = tab_line-d
e = tab_line-e ) ).
"Example using a FOR loop with the REDUCE operator and LET
DATA(itab10) = REDUCE tt_type3( INIT tab = VALUE #( )
FOR wa5 IN itab1
LET tableline = VALUE #( itab2[ key2 = wa5-key1 ] OPTIONAL ) IN
NEXT tab = VALUE #( BASE tab
( key1 = wa5-key1
a = wa5-a
b = wa5-b
d = tableline-d
e = tableline-e ) ) ).
- Sorted tables are stored in the memory in an automatically sorted
order, hence, they cannot be sorted explicitly with
SORT
. - For standard and hashed tables, the order can be changed.
- When using
SORT
statements, the sort order is derived either by the primary table key (Note: Secondary keys cannot be used for the sorting.) or by explicitly specifying the fields to be sorted by. - Explicit specification is the recommended way because it is easier to understand and can prevent unwanted sorting results, especially with tables with standard key.
- You can also sort dynamically. For more information, refer to the Dynamic Programming cheat sheet.
Subject | Details/Code Snippet |
Sorting by primary table key |
"Implicit sorting by primary table key and in ascending order by default
SORT itab.
"Optional additions to determine the sort order
"As mentioned above, ASCENDING is used implicitly. Here, specifying it explicitly.
SORT itab ASCENDING.
SORT itab DESCENDING. The effect of sorting can have an unexpected result if you use the simple form of the statement and do not explicitly specify the keys. If an internal table has a structured line type and (perhaps inadvertently) the standard key as the primary table key, that is, all character-like and byte-like components make up the primary table key, all these components are taken into account when the table is sorted. "Is basically the same as it2
DATA it1 TYPE TABLE OF zdemo_abap_fli.
DATA it2 TYPE STANDARD TABLE OF zdemo_abap_fli WITH DEFAULT KEY.
"Respecting the standard key when sorting
SORT it1. Plus: Suppose there are only elementary numeric components in an internal table with a structured line type. In this case, sorting has no effect because the primary table key is considered empty. This is certainly also true for tables declared with |
Sorting by explicitly specifying components |
You can sort by any component of the internal table. It is also possible to specify the sort order
(even component-wise). Explicitly specifying the components has the advantage that your code is easier to understand and you can avoid unexpected results if you accidentally use DATA it3 TYPE TABLE OF struc WITH NON-UNIQUE KEY a.
"Sorting by primary table key a
SORT itab.
"Specifying the component to sort for; here, it is the same as the key;
"this way, the sorting is easier to understand
SORT itab BY a.
"Syntax showing multiple component sorting with component-wise sort order
SORT itab BY a b ASCENDING c DESCENDING.
"Sorting respecting the entire line (e. g. in the context of tables with
"empty or standard keys)
SORT itab BY table_line. |
As mentioned above, you can modify the content of internal table lines directly in the context of READ TABLE
and LOOP AT
statements using field symbols and data reference variables. You can also use table expressions for direct modification (as also covered in section Table Expressions). Note that the key fields of the primary table key of sorted and hashed tables are always read-only. If you try to modify a key field, a runtime error occurs. However, this is not checked until runtime.
The following examples demonstrate direct modification of recently read table lines:
"Declaring and populating demo internal tables
TYPES: BEGIN OF ty_struc,
comp1 TYPE i,
comp2 TYPE string,
comp3 TYPE c LENGTH 3,
END OF ty_struc.
DATA it_st TYPE TABLE OF ty_struc WITH NON-UNIQUE KEY comp1.
DATA it_so TYPE SORTED TABLE OF ty_struc WITH UNIQUE KEY comp1.
DATA it_ha TYPE HASHED TABLE OF ty_struc WITH UNIQUE KEY comp1.
it_st = VALUE #( ( comp1 = 1 comp2 = `AAAAAA` comp3 = 'bbb' )
( comp1 = 2 comp2 = `CCCCCC` comp3 = 'ddd' )
( comp1 = 3 comp2 = `EEEEEE` comp3 = 'fff' )
( comp1 = 4 comp2 = `GGGGGG` comp3 = 'hhh' )
).
it_so = it_st.
it_ha = it_st.
"---- Modifying internal table content by changing the ----
"---- content of READ TABLE statement target areas --------
"Reading table line into a target area
READ TABLE it_st INTO DATA(wa) INDEX 1.
READ TABLE it_so ASSIGNING FIELD-SYMBOL(<fs>) INDEX 2.
READ TABLE it_ha REFERENCE INTO DATA(dref) WITH TABLE KEY comp1 = 3. "No reading by index in case of hashed tables
"------ Modification examples -------
"Modifying all non-key components using the VALUE operator and
"the BASE addition
<fs> = VALUE #( BASE <fs> comp2 = `IIIIII` comp3 = 'jjj' ).
"In the following example, the key value is assigned a new
"value. Key values are protected against change in case of key tables.
"A runtime error occurs.
"<fs> = VALUE #( comp1 = 5 comp2 = `IIIIII` comp3 = 'jjj' ).
dref->* = VALUE #( BASE dref->* comp2 = `KKKKKK` comp3 = 'lll' ).
"Same as above. Key values cannot be changed in this case.
"dref->* = VALUE #( comp1 = 5 comp2 = `MMMMMM` comp3 = 'nnn' ).
"Using a MODIFY statement outlined below for changing internal
"table content based on a read line in a work area
MODIFY TABLE it_st FROM VALUE #( BASE wa comp2 = `OOOOOO` comp3 = 'ppp' ).
"Modifying individual components
READ TABLE it_st INTO wa INDEX 2.
READ TABLE it_so ASSIGNING <fs> INDEX 3.
READ TABLE it_ha REFERENCE INTO dref WITH TABLE KEY comp1 = 4.
"Using VALUE/BASE
<fs> = VALUE #( BASE <fs> comp2 = `QQQQQQ` ).
dref->* = VALUE #( BASE dref->* comp2 = `RRRRRR` ).
MODIFY TABLE it_st FROM VALUE #( BASE wa comp2 = `SSSSSS` ).
"Using the component selector
<fs>-comp3 = 'ttt'.
READ TABLE it_st INTO wa INDEX 3.
wa-comp3 = 'uuu'.
MODIFY TABLE it_st FROM wa.
"Object component selector in case of dereferencing ...
dref->comp2 = `VVVVVV`.
"... which is a more comfortable option compared to using the
"dereferencing and component selector operators in the following way.
dref->*-comp3 = 'www'.
"---- Modifying internal table content using table expressions -----
"Changing the entire table line of a standard table
"In standard tables, the key value change is allowed.
it_st[ 3 ] = VALUE #( comp1 = 9 comp2 = `XXXXXX` comp3 = 'yyy' ).
"As above, the sorted table is a key table having a unique key,
"therefore a write cannot be performed on the entire entry. Runtime
"errors can occur.
"it_so[ 3 ] = VALUE #( comp2 = `XXXXXX` comp3 = 'yyy' ).
"The same applies to hashed tables.
"it_ha[ comp2 = `OOOOOO` ] = VALUE #( comp2 = `XXXXXX` comp3 = 'yyy' ).
"Changing individual components
it_st[ 3 ]-comp2 = `ZZZZZZ`.
it_so[ 3 ]-comp3 = 'A1'.
it_ha[ comp2 = `CCCCCC` ]-comp2 = `B2`.
"As above, no key field change in key tables. Allowed in standard
"tables.
"it_so[ 3 ]-comp1 = 10.
"it_ha[ comp2 = `AAAAAA` ]-comp1 = `C3`.
it_st[ 1 ]-comp1 = 99.
"---- Modifying table content in all table rows in a loop ----
"For more syntax options regarding loops, check the section above.
"Target area: field symbol
LOOP AT it_st ASSIGNING FIELD-SYMBOL(<lo>).
<lo>-comp2 = sy-tabix.
ENDLOOP.
"---- Modifying table content restricting the rows that are looped across ----
"Target area: data reference variable
LOOP AT it_st reference into data(lo) FROM 2 TO 3.
lo->comp3 = sy-tabix.
ENDLOOP.
"Target area: work area
LOOP AT it_so into data(wa_lo) where comp1 < 4.
wa_lo-comp2 = sy-tabix.
MODIFY TABLE it_so FROM wa_lo.
ENDLOOP.
MODIFY
statements provide multiple ways of changing the content of single and multiple table lines by specifying the table key or a table index,
without first reading the lines into a target area.
"Addition FROM ...; specified key values determine the line to be modified
"line: existing line including key values
MODIFY TABLE it FROM line.
"line constructed inline
MODIFY TABLE it FROM VALUE #( a = 1 b = 2 ... ).
"Respecting only specified fields with the addition TRANSPORTING
"In case of sorted/hashed tables, key values cannot be specified.
MODIFY TABLE it FROM line TRANSPORTING b c.
"Modification via index
"Note that it is only MODIFY, not MODIFY TABLE.
"Example: It modifies the line with number 1 in the primary table index.
MODIFY it FROM line INDEX 1.
"Without the addition TRANSPORTING, the entire line is changed.
"Example: It modifies specific values.
MODIFY it FROM line INDEX 1 TRANSPORTING b c.
"USING KEY addition
"If the addition is not specified, the primary table key is used;
"otherwise, it is the explicitly specified table key that is used.
"Example: It is the same as MODIFY it FROM line INDEX 1.
MODIFY it FROM line USING KEY primary_key INDEX 1.
"The statement below uses a secondary key and an index specification
"for the secondary table index. Only specific fields are modified.
MODIFY it FROM line USING KEY sec_key INDEX 1 TRANSPORTING c d.
"Modifying multiple lines in internal tables
"All lines matching the logical expression in the WHERE clause are modified
"as specified in line.
"The additions TRANSPORTING and WHERE are both mandatory; USING KEY is optional.
MODIFY it FROM line TRANSPORTING b c WHERE a < 5.
💡 Note
- The system field
sy-subrc
is set to0
if at least one line was changed. It is set to4
if no lines were changed.MODIFY
,DELETE
, andINSERT
statements can be specified with and without theTABLE
addition. WithTABLE
means an index access. WithoutTABLE
means an access via the table key.
You can use DELETE
statements to delete single and multiple lines in internal tables. The following additions can be used: USING KEY
(for specifying a table key), FROM
/TO
(for specifying row ranges), STEP
(for specifying the step size), and WHERE
(for specifying conditions).
"-------------- Deleting via index --------------
"Example: The first line in the table is deleted.
DELETE it INDEX 1.
"If USING KEY is not used, INDEX can only be used with index tables.
"If doing so, it determines the line from the primary table index.
"If a secondary key is specified, the secondary table index is respected
"Example: same as above
DELETE it INDEX 1 USING KEY primary_key.
"Deleting an index range; FROM or TO alone can also be specified
DELETE it FROM 2 TO 5.
"-------------- Deleting via keys --------------
"The line must have a compatible type to the tables line type and
"include key values. The first found line with the corresponding keys
"is deleted.
"If the key is empty, no line is deleted.
DELETE TABLE it FROM line.
"Instead of specifying the keys using a data object ("line" above),
"the keys can be specified separately. All key values must be specified.
"Example: Respects keys from primary table index.
DELETE TABLE it WITH TABLE KEY a = 1.
"You can also specify secondary keys.
"Example: Same as above
DELETE TABLE it WITH TABLE KEY primary_key COMPONENTS a = 1.
DELETE TABLE it_sec WITH TABLE KEY sec_key COMPONENTS ...
"---------- Deleting multiple lines based on a WHERE condition ----------
"Specifying the additions USING KEY, FROM, TO is also possible.
DELETE it WHERE a < 6.
"Excursion: Deleting in a LIKE-like fashion you may know from
"ABAP SQL statements.
"The LIKE addition is not available for the WHERE clause in DELETE
"statements for internal tables as is the case for ABAP SQL DELETE statements.
DATA(str_table) = VALUE string_table( ( `abcZ` ) ( `Zdef` ) ( `gZhi` )
( `Zjkl` ) ( `Zmno` ) ( `pqrZ` ) ).
"You can, for example, use logical operators such as CP (conforms to pattern)
"All lines that begin with Z are to be deleted.
DELETE str_table WHERE table_line CP `Z*`.
"Result: abcZ / gZhi / pqrZ
"---------- Deleting the current line inside a LOOP statement ----------
"The following example illustrates deleting the current table line
"using a DELETE statement within a LOOP statement. Lines with even
"numbers are deleted.
"Note:
"- The short form of the DELETE statement always deletes the
" current first line implicitly. It is only possible within a LOOP
" statement and the delete operation is performed on the same internal
" table.
"- The field symbol (or reference variable) should not be used after
" the DELETE statement any more.
DATA itab1 TYPE TABLE OF i WITH EMPTY KEY.
itab1 = VALUE #( ( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 ) ( 7 ) ( 8 ) ( 9 ) ( 10 ) ).
LOOP AT itab1 ASSIGNING FIELD-SYMBOL(<fs>).
IF <fs> MOD 2 = 0.
DELETE itab1.
ENDIF.
ENDLOOP.
*Table content:
*1
*3
*5
*7
*9
"The following, similar example (uneven numbers are deleted) uses a
"table which is looped over by specifying the addition USING KEY.
"In this case (using LOOP ... USING KEY ...), the short form of the
"DELETE statement cannot be used. Use the DELETE statement with the
"addition USING KEY loop_key to delete the current first line.
"loop_key is a predefined name to be used with DELETE and within
"loops that specify LOOP ... USING KEY .... No other key name is
"possible here.
DATA itab2 TYPE TABLE OF i WITH NON-UNIQUE KEY table_line.
itab2 = VALUE #( ( 1 ) ( 2 ) ( 3 ) ( 4 ) ( 5 ) ( 6 ) ( 7 ) ( 8 ) ( 9 ) ( 10 ) ).
LOOP AT itab2 USING KEY primary_key REFERENCE INTO DATA(dref2).
IF dref2->* MOD 2 <> 0.
DELETE itab2 USING KEY loop_key.
ENDIF.
ENDLOOP.
*Table content:
*2
*4
*6
*8
*10
DELETE ADJACENT DUPLICATES
statements allow you to delete all adjacent lines except for the first line that have the same content in certain components. You usually need to perform some appropriate sorting before using these statements.
"Implicitly using the primary table key
DELETE ADJACENT DUPLICATES FROM it.
"Deletion respecting the values of the entire line
DELETE ADJACENT DUPLICATES FROM it COMPARING ALL FIELDS.
"Only lines are deleted with matching content in specific fields
DELETE ADJACENT DUPLICATES FROM it COMPARING a c.
"Deletion respecting a specified table key
"Same as first example above
DELETE ADJACENT DUPLICATES FROM it USING KEY primary_key.
DELETE ADJACENT DUPLICATES FROM it USING KEY sec_key.
💡 Note
The system fieldsy-subrc
is set to0
if at least one line has been deleted. It is set to4
if no lines were deleted.
The
CLEAR
and
FREE
statements allow you to delete the entire table content.
The difference between the two is in the handling of the memory space originally allocated to the table. When a table is cleared with CLEAR
,
the content is removed, but the memory space initially requested remains
allocated. If the table is populated again later, the memory space is still
available, which is a performance advantage over
clearing an internal table with FREE
. Such a statement also
deletes the table content, but it also releases the memory
space.
Note that an assignment using the VALUE
operator without entries in the parentheses clears the internal table.
CLEAR it.
"This statement additionally releases memory space.
FREE it.
"Assignment using the VALUE operator without entries in the parentheses
it = VALUE #( ).
"The same applies to data reference variables pointing to internal tables.
it_ref = NEW #( ).
To group internal tables, there are additions for LOOP AT
statements.
LOOP AT ... GROUP BY
-
Groups the lines of internal tables based on a group key. Each group key represents a group, and the lines read are members of that group.
-
Performs a loop across these groups.
-
You can execute a nested loop across the members of each group using
LOOP AT GROUP
statements. -
The statements offer a variety of syntax options, such as the following. For more information, refer to the ABAP Keyword Documentation:
ASCENDING
/DESCENDING
additions to specify the sort order in the group loop. You can also sort beforehand withSORT
statements.WITHOUT MEMBERS
: This creates groups (requiring group key binding) without assigning actual component values. Access to group lines is not possible. Use this when access is not necessary and you want to enhance read performance.- Specifying the group key after
GROUP BY
- This can be a single data object or multiple keys within parentheses, which define a structure with specific components.
- In simple cases, component values of the table line can be assigned to the group key or the components of the group key. However, expressions are also possible on the right side of the group key assignments.
- Storing group-specific information in structured group keys.
GROUP SIZE
to count group members, e.g.,gs = GROUP SIZE
.GROUP INDEX
to index group members, e.g.,gi = GROUP INDEX
.- You can choose component names (such as
gs
andgi
in the previous example) freely. - To use these components, you need a group key binding.
- Group key binding
- Can be specified at the end of the statement using
INTO
and a data object. You can also use data references (REFERENCE INTO
) or field symbols (ASSIGNING ...
). - If specified, the current key's group key is assigned to the specified data object (data reference, or field symbol). You can then address the group in nested loops with
LOOP AT GROUP
. - Note: You can access the read result (e.g.,
dobj
inLOOP AT itab INTO dobj
), but the component values are initial when the group key binding is specified. You can address the group and its component values in nested loops withLOOP AT GROUP
. - If you do not specify the group key binding, it defaults to a representative binding. In each loop pass, the first line of the current group is assigned to the target area. You can process this representative further in nested loops with
LOOP AT GROUP
. With representative binding, thesy-tabix
value is set as if the loop was specified without grouping.
- Can be specified at the end of the statement using
LOOP AT GROUP
:
- Allows for a nested loop across group members.
- Only applicable with
LOOP ... GROUP BY
statements (providedWITHOUT MEMBERS
is not specified) - After using
LOOP AT GROUP
, you can specify ...- the read result that serves as a representative in representative binding:
LOOP AT it INTO DATA(wa) GROUP BY ... LOOP AT GROUP wa ...
. - the group key binding:
LOOP AT it INTO DATA(wa) GROUP BY ... INTO DATA(gkb). ... LOOP AT GROUP gkb ...
.
- the read result that serves as a representative in representative binding:
- Additional syntax options like a
WHERE
condition and further grouping are also available.
More information:
- ABAP Keyword Documentation
- Iteration expressions can also handle table grouping (
FOR ... IN GROUP
). For example, see the Constructor Expressions cheat sheet. - Internal Tables: Grouping cheat sheet
The example class below demonstrates internal table grouping options. To try it out, create a demo class named zcl_some_class
and paste the following code into it. After activation, choose F9 in ADT to execute the class. The example is designed to display results in the console.
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
TYPES: BEGIN OF demo_struct,
comp1 TYPE c LENGTH 1,
comp2 TYPE i,
comp3 TYPE abap_boolean,
comp4 TYPE string,
END OF demo_struct,
tab_type TYPE TABLE OF demo_struct WITH EMPTY KEY.
DATA str_table TYPE string_table.
"Populating a demo internal table as the basis of the syntax example
"Note: The example loops only use data objects as targets, not data references
"or field symbols.
DATA(it) = VALUE tab_type( ( comp1 = 'd' comp2 = 0 comp3 = abap_false )
( comp1 = 'a' comp2 = 1 comp3 = abap_true )
( comp1 = 'a' comp2 = 2 comp3 = abap_false )
( comp1 = 'e' comp2 = 11 comp3 = abap_true )
( comp1 = 'b' comp2 = 5 comp3 = abap_true )
( comp1 = 'b' comp2 = 6 comp3 = abap_false )
( comp1 = 'a' comp2 = 3 comp3 = abap_false )
( comp1 = 'b' comp2 = 4 comp3 = abap_true )
( comp1 = 'c' comp2 = 10 comp3 = abap_true )
( comp1 = 'e' comp2 = 1 comp3 = abap_false )
( comp1 = 'd' comp2 = 7 comp3 = abap_true )
( comp1 = 'a' comp2 = 4 comp3 = abap_true )
( comp1 = 'e' comp2 = 111 comp3 = abap_true ) ).
"The following example (and several others below) does not specify a nested loop.
"It does not specify a group key binding either. This means that the work area
"contains the first line of each group, representing the group in the loop
"(representative binding). The comp4 component is assigned the sy-tabix value,
"which is the number of the line in the table without the grouping.
DATA ita LIKE it.
LOOP AT it INTO DATA(waa) GROUP BY waa-comp1.
waa-comp4 = sy-tabix.
APPEND waa TO ita.
ENDLOOP.
out->write( data = ita name = `ita` ).
"Specifying sort order
DATA itb LIKE it.
LOOP AT it INTO DATA(wab) GROUP BY wab-comp1 ASCENDING.
wab-comp4 = sy-tabix.
APPEND wab TO itb.
ENDLOOP.
out->write( data = itb name = `itb` ).
"WITHOUT MEMBERS addition; a group key binding is required
"after WITHOUT MEMBERS
"The group key binding is added to a string table for visualizing its
"content.
"Note: The component values are initial when the group key binding is
"specified.
LOOP AT it INTO DATA(wac) GROUP BY wac-comp1 WITHOUT MEMBERS INTO DATA(keyc).
ASSERT wac IS INITIAL.
APPEND keyc TO str_table.
ENDLOOP.
out->write( data = str_table name = `str_table` ).
"Using a structured group key
"The following example just assigns component values to the group key. In this case,
"the grouping is performed with more than just one criterion as in the previous examples.
"As a result, table lines are added to the other table in descending order based on the
"two component values.
DATA itd LIKE it.
LOOP AT it INTO DATA(wad) GROUP BY ( key1 = wad-comp1 key2 = wad-comp2 ) DESCENDING.
APPEND wad TO itd.
ENDLOOP.
out->write( data = itd name = `itd` ).
"In the following example, the group is sorted in ascending order. Note that the
"group index value uses the original position in the group index. The group key
"binding information is added to a string table for visualizing its content.
CLEAR str_table.
LOOP AT it INTO DATA(wae) GROUP BY ( key = wae-comp1 gi = GROUP INDEX gs = GROUP SIZE ) ASCENDING INTO DATA(keye).
APPEND |Key component: '{ keye-key }', group index: '{ keye-gi }', group size: '{ keye-gs }'| TO str_table.
ENDLOOP.
out->write( data = str_table name = `str_table` ).
"LOOP AT GROUP: Nested loop across group members
"Unlike the previous example, the example uses a nested loop across the groups (the group key binding is
"specified after LOOP AT GROUP). There, the component values of the members can be accessed.
DATA itf LIKE it.
LOOP AT it INTO DATA(waf) GROUP BY ( key = waf-comp1 gi = GROUP INDEX gs = GROUP SIZE ) ASCENDING INTO DATA(keyf).
LOOP AT GROUP keyf INTO DATA(memberf).
APPEND VALUE #( comp1 = memberf-comp1 comp2 = memberf-comp2 comp3 = memberf-comp3
comp4 = |Key component: '{ keyf-key }', group index: '{ keyf-gi }', group size: '{ keyf-gs }'|
) TO itf.
ENDLOOP.
ENDLOOP.
out->write( data = itf name = `itf` ).
"The objective of this example is to extract the line with the highest value in a particular
"column within a group from the original table to another.
"The example uses representative binding, i.e. the representative of the group is specified
"in the work area, not in a group key binding.
DATA itg LIKE it.
LOOP AT it INTO DATA(wag) GROUP BY wag-comp1 ASCENDING.
LOOP AT GROUP wag INTO DATA(memberg) GROUP BY memberg-comp2 DESCENDING.
APPEND memberg TO itg.
EXIT.
ENDLOOP.
ENDLOOP.
out->write( data = itg name = `itg` ).
"The following example is similar to the previous example, and yields the same result.
"Here, the group key binding is specified after LOOP AT GROUP.
DATA ith LIKE it.
LOOP AT it INTO DATA(wah) GROUP BY wah-comp1 ASCENDING.
LOOP AT GROUP wah INTO DATA(memberh) GROUP BY memberh-comp2 DESCENDING.
APPEND memberh TO ith.
EXIT.
ENDLOOP.
ENDLOOP.
ASSERT itg = ith.
out->write( data = ith name = `ith` ).
"Additional syntax options, like specifying a WHERE condition in both nested and outer
"loops, are possible. The example below shows that the LOOP AT GROUP statement assigns
"the value of sy-tabix to the value that would be set for the current line in the LOOP
"without grouping.
DATA iti LIKE it.
LOOP AT it INTO DATA(wai) GROUP BY wai-comp1 ASCENDING.
LOOP AT GROUP wai INTO DATA(memberi) WHERE comp3 = abap_true.
APPEND VALUE #( comp1 = memberi-comp1 comp2 = memberi-comp2 comp3 = memberi-comp3
comp4 = |sy-tabix: '{ sy-tabix }'|
) TO iti.
ENDLOOP.
ENDLOOP.
out->write( data = iti name = `iti` ).
ENDMETHOD.
ENDCLASS.
- You can use
COLLECT
statements, for example, to add the values of numeric components to the corresponding values in an internal table. - It is recommended that you use it mainly for internal tables with a unique primary key, especially hashed tables.
- Find more information here
"This example demonstrates how to insert data from a database table
"into an internal table in a compressed way. Within a SELECT loop,
"a COLLECT statement is used to consolidate lines with identical
"primary key components (carrid and connid) by summing the number
"of occupied seats in the numeric component (seatsocc).
"Additionally, an internal table is filled by adding all read lines.
"This table is looped across to simulate the effect of the COLLECT
"statement.
DATA: BEGIN OF seats,
carrid TYPE zdemo_abap_fli-carrid,
connid TYPE zdemo_abap_fli-connid,
seatsocc TYPE zdemo_abap_fli-seatsocc,
END OF seats,
seats_tab_col LIKE HASHED TABLE OF seats WITH UNIQUE KEY carrid connid,
seats_tab_all LIKE TABLE OF seats WITH EMPTY KEY,
seats_tab_loop_grp LIKE seats_tab_col.
SELECT carrid, connid, seatsocc
FROM zdemo_abap_fli
INTO @seats.
COLLECT seats INTO seats_tab_col.
APPEND seats TO seats_tab_all.
ENDSELECT.
LOOP AT seats_tab_all INTO DATA(wa) GROUP BY ( key1 = wa-carrid key2 = wa-connid ).
INSERT VALUE #( carrid = wa-carrid connid = wa-connid ) INTO TABLE seats_tab_loop_grp ASSIGNING FIELD-SYMBOL(<fs>).
LOOP AT GROUP wa INTO DATA(member).
<fs>-seatsocc = <fs>-seatsocc + member-seatsocc.
ENDLOOP.
ENDLOOP.
ASSERT seats_tab_loop_grp = seats_tab_col.
Consider a scenario where you have a standard internal table, and you frequently access its content using a free key. The table is declared without a secondary table key. You can add a secondary table key to improve read performance.
The following example creates two demo internal tables. One without a secondary table key and the other with a secondary table key. The tables are populated with a lot of data. Then, in a DO
loop, many reads are performed on the internal tables. One example uses a free key for the read, the other uses a secondary table key that includes the components used for the free key search. Before and after the reads, the current timestamp is stored in variables, from which the elapsed time is calculated. There should be a significant delta of the elapsed time.
CLASS zcl_some_class DEFINITION PUBLIC FINAL CREATE PUBLIC.
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
TYPES: BEGIN OF demo_struc,
idx TYPE i,
str TYPE string,
num TYPE i,
END OF demo_struc.
DATA itab TYPE TABLE OF demo_struc WITH NON-UNIQUE KEY idx.
DATA itab_sec TYPE TABLE OF demo_struc
WITH NON-UNIQUE KEY idx
WITH NON-UNIQUE SORTED KEY sk
COMPONENTS str num.
DATA runtime_tab TYPE TABLE OF decfloat34 WITH EMPTY KEY.
"Constant values on the basis of which the test runs are performed
CONSTANTS: num_of_table_lines TYPE i VALUE 5000,
num_of_repetitions TYPE i VALUE 10,
num_of_reads TYPE i VALUE 1000.
"Populating demo tables
DO num_of_table_lines TIMES.
INSERT VALUE #( idx = sy-index
str = |INDEX{ sy-index }|
num = sy-index ) INTO TABLE itab.
ENDDO.
itab_sec = itab.
"To get a meaningful result, many read iterations are performed (as defined
"by 'num_of_reads'). These iterations are performed multiple times as indicated
"by 'num_of_repetitions'. The current timestamp is stored before and after the
"read iterations. The difference between these two timestamps provides the
"duration, in seconds, of how long the reads took. This value is stored in
"an internal table. The fastest value is taken as reference value for the
"comparison with the value from the second internal table below.
"Reading from an internal table using a free key
DO num_of_repetitions TIMES.
DATA(ts1) = utclong_current( ).
DO num_of_reads TIMES.
"The free key corresponds to the secondary table key specified for the
"table in the second example.
READ TABLE itab WITH KEY str = `INDEX` && sy-index num = sy-index TRANSPORTING NO FIELDS.
ENDDO.
DATA(ts2) = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = DATA(seconds) ).
APPEND seconds TO runtime_tab.
ENDDO.
SORT runtime_tab BY table_line ASCENDING.
out->write( `Elapsed time for the reads using a free key:` ).
DATA(fastest_free_key) = runtime_tab[ 1 ].
out->write( fastest_free_key ).
out->write( repeat( val = `-` occ = 70 ) ).
CLEAR runtime_tab.
"Reading from an internal table using the secondary table key
DO num_of_repetitions TIMES.
ts1 = utclong_current( ).
DO num_of_reads TIMES.
READ TABLE itab_sec WITH TABLE KEY sk COMPONENTS str = `INDEX` && sy-index num = sy-index TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
APPEND seconds TO runtime_tab.
ENDDO.
SORT runtime_tab BY table_line ASCENDING.
out->write( `Elapsed time for the reads using a secondary table key:` ).
DATA(fastest_sec_key) = runtime_tab[ 1 ].
out->write( fastest_sec_key ).
out->write( |\n\n| ).
DATA(percentage) = fastest_sec_key / fastest_free_key * 100.
out->write( |In the test runs of this example, the fastest read access with the secondary table key takes approximately | &&
|{ percentage DECIMALS = 2 }% of the time it takes for the fastest read using a free key.| ).
ENDMETHOD.
ENDCLASS.
- The following example explores the performance of read accesses to internal tables of various kinds in a simplified and self-contained manner.
- It primarily addresses the scenario of large internal tables that are accessed frequently.
- Read accesses are executed using
READ TABLE
statements, which access the tables by:- Index (both primary and secondary table index)
- Key (primary table key, secondary table key, free key)
- To try this, create a demo class named
zcl_some_class
and insert the provided code. After activation, choose F9 in ADT to execute the class. It may take some time to finish and display the output. The example is set up to display the results of the read performance test in the console.
💡 Note
- This example is used for exploration and experimentation
⚠️ . It is solely for demonstration purposes, and it is not a tool for proper and accurate runtime and performance testing. Due to its simplified nature, the results may not be entirely accurate. However, multiple test runs should reflect the notes below.- The example concentrates on a few demo internal tables, constructed using various declaration options.
- The purpose of this example is to underscore the significance of choosing the right table categories for your internal tables, tailored to your specific use case and the frequency of table access.
Notes on ...
... primary table key and primary table index:
- Each internal table has a primary key, which can be custom-defined or use the standard key. In standard tables, it is also possible to use an empty key.
- Be aware of the implications of specifying the standard key, either intentionally or unintentionally. The standard key combines all character-like components, such as
txt
in the example below, while the remaining components are of typei
. - The primary key of a table is ...
- a hash key in hashed tables,
- a sorted key in sorted tables,
- always non-unique in standard tables.
- Be aware of the implications of specifying the standard key, either intentionally or unintentionally. The standard key combines all character-like components, such as
- Key access is optimized for hashed and sorted tables, but not for standard tables.
- You cannot modify key fields in hashed and sorted tables.
- In standard tables, key access is not optimized regarding the primary table index (for the secondary table index, it is optimized) because the search is linear.
- Generally, the larger the table, the slower the key access. The benefit of hash tables is that they offer constant access time, even for very large tables.
- Sorted and standard tables have a primary table index, hence their designation as index tables.
- The primary table index assigns a unique line number to each table line.
- The index updates whenever a line is added or removed.
- Unlike key access, access time via the index does not increase linearly with table size.
... secondary table key and secondary table index:
- Secondary table keys, which can be sorted or hashed, are available for all table categories.
- They enhance table access efficiency and performance.
- Declaring a secondary table key generates a corresponding secondary table index. However, the index for a non-unique key does not update immediately upon adding or deleting a line. The update happens when the internal table is accessed using the secondary table key.
- In ABAP statements, you must specify the secondary table key explicitly. Otherwise, the primary table key is used implicitly.
- Data access using the secondary table key is always optimized, even for standard tables. Thus, even older standard tables can gain from optimized access by adding secondary table keys later, without impacting existing table-related statements.
- However, weigh the administrative costs of secondary table keys. Use them only in scenarios where they offer substantial benefits, like large internal tables that are filled once and rarely altered. Frequent modifications can lead to regular index updates, potentially impacting performance.
... the use of table categories:
-
Standard tables:
- Suitable for ...
- small, sequentially accessed tables that are often accessed by index
- tables where sorting is not critical. However, you can explicitly sort using
SORT
statements, especially after table population. - tables that are populated frequently, as there is no need to check for unique entries regarding the primary table key.
- Access speed:
- Fast: By index, optimized key access with secondary table keys, and free keys using
READ TABLE ... BINARY SEARCH
statements (Note: It is recommended to use secondary table keys for an optimized access.) - Slow: Primary table and free key
- Fast: By index, optimized key access with secondary table keys, and free keys using
- Suitable for ...
-
Sorted tables:
- Suitable for ...
- large tables requiring consistent sorted content
- tables frequently accessed both by index and in sequence
- Access is fast by index, primary table key (always optimized), secondary table key (also optimized)
- Suitable for ...
-
Hashed tables:
- Suitable for ...
- very large tables that are filled once and rarely altered
- tables where key-based access is the primary method
- tables that do not require index access on the primary table index
- Access is ...
- fast and optimized for both primary and secondary table keys
- consistent for large internal tables due to a special hash algorithm
- Suitable for ...
Expand the following collapsible section for example code. To try it out, create a demo class named zcl_some_class
and paste the code into it. After activation, choose F9 in ADT to execute the class. Note that, when running the class, it may take a while to complete and display output in the console.
🟢 Click to expand for example code
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
PROTECTED SECTION.
PRIVATE SECTION.
"Internal table to process demo tables and storing information
TYPES: BEGIN OF info_struc,
id TYPE i,
name TYPE string,
runtime TYPE decfloat34,
operation TYPE string,
comment TYPE string,
lines TYPE i,
accesses TYPE i,
table_kind TYPE abap_tablekind,
keys TYPE abap_table_keydescr_tab,
itab_ref TYPE REF TO data,
END OF info_struc,
info_tab_type TYPE TABLE OF info_struc WITH EMPTY KEY.
DATA info_tab TYPE TABLE OF info_struc WITH EMPTY KEY.
"Object reference variable to retrieve type information
"at runtime (RTTI)
DATA tdo TYPE REF TO cl_abap_tabledescr.
"Demo internal tables
"The line type is used for all demo tables.
TYPES: BEGIN OF s,
idx TYPE i,
txt TYPE c LENGTH 20,
num TYPE i,
END OF s.
"------------------ Standard tables ------------------
"Empty primary table key
DATA it_std_empty_key TYPE TABLE OF s WITH EMPTY KEY.
"Primary table key explicitly specified (only non-unique in case of standard tables)
DATA it_std_w_nu_pr_key TYPE TABLE OF s WITH NON-UNIQUE KEY idx.
"Secondary table key specified, primary key not specified explicitly
"Therefore, the standard primary table key is used. It consists of all
"character-like components (which is the 'txt' component only in this case).
DATA it_std_w_std_pr_key_w_sec_key TYPE TABLE OF s WITH NON-UNIQUE SORTED KEY sec_key COMPONENTS num.
"Note: The following declaration corresponds to the previous one. Here, the standard key
"is specified explicitly.
"DATA it_std_w_std_pr_key_w_sec_key TYPE TABLE OF s WITH DEFAULT KEY WITH NON-UNIQUE SORTED KEY sec_key COMPONENTS num.
"Empty primary table key, secondary table key specified
DATA it_std_w_emp_pr_key_w_sec_key TYPE TABLE OF s WITH EMPTY KEY WITH NON-UNIQUE SORTED KEY sec_key COMPONENTS num.
"------------------ Sorted tables ------------------
DATA it_sorted TYPE SORTED TABLE OF s WITH UNIQUE KEY idx.
"Secondary table key specified
DATA it_sorted_w_sec_key TYPE SORTED TABLE OF s WITH UNIQUE KEY idx WITH NON-UNIQUE SORTED KEY sec_key COMPONENTS num.
"------------------ Hashed tables ------------------
DATA it_hashed TYPE HASHED TABLE OF s WITH UNIQUE KEY idx.
"Secondary table key specified
DATA it_hashed_w_sec_key TYPE HASHED TABLE OF s WITH UNIQUE KEY idx WITH NON-UNIQUE SORTED KEY sec_key COMPONENTS num.
"------------------ Method declarations ------------------
"Populating the internal tables with demo data
"The method is supplied with the number of lines that should be added to the demo tables. Ideally, and to
"demonstrate a meaningful result, it is a high integer value. Furthermore, the information table is
"prepared for output purposes. It also contains a reference to the internal table content that is processed.
METHODS prepare_itabs IMPORTING number_of_lines TYPE i.
"For performing the read accesses
"The method is supplied with integer values.
"- Number of read accesses: Ideally, and to demonstrate a meaningful result, it is a high integer value (but not
" higher than the number of internal table lines in the example).
"- Number of repetitions: The higher the number of repetitions (i.e. how often the read access is performed n times),
" the more accurate the performance check can be. In the result, the fastest read access time is stored in the
" information table, which is returned.
METHODS read_itabs IMPORTING number_of_reads TYPE i
number_of_repetitions TYPE i
RETURNING VALUE(read_results) TYPE info_tab_type.
"Helper method to check that specific components are included in the internal
"tables since READ statements are performed using these components.
METHODS check_example_components RETURNING VALUE(are_included) TYPE abap_boolean.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
prepare_itabs( number_of_lines = 5000 ).
"Specifying a fairly high number of read accesses to the internal tables,
"and not just one or two repetitions of the runtime checks so as to get
"a more precise result (the fastest access is returned and displayed in the
"output table). The higher the numbers provided, the longer the example
"runs.
DATA(read_result) = read_itabs( number_of_reads = 1000
number_of_repetitions = 20 ).
out->write( read_result ).
ENDMETHOD.
METHOD prepare_itabs.
"Populating internal tables with demo data
DO number_of_lines TIMES.
INSERT VALUE #( idx = sy-index txt = |INDEX{ sy-index }| num = sy-index ) INTO TABLE it_std_empty_key.
ENDDO.
"Copying the content to the other internal tables having the same line type
"to have the same data basis for the runtime checks
it_std_w_nu_pr_key = it_std_empty_key.
it_std_w_std_pr_key_w_sec_key = it_std_empty_key.
it_std_w_emp_pr_key_w_sec_key = it_std_empty_key.
it_sorted = it_std_empty_key.
it_sorted_w_sec_key = it_std_empty_key.
it_hashed = it_std_empty_key.
it_hashed_w_sec_key = it_std_empty_key.
"Populating the information table (including references to the tables)
info_tab = VALUE #(
( name = `it_std_empty_key` itab_ref = REF #( it_std_empty_key ) lines = lines( it_std_empty_key ) comment = `Standard, empty primary t.key` )
( name = `it_std_w_nu_pr_key` itab_ref = REF #( it_std_w_nu_pr_key ) lines = lines( it_std_w_nu_pr_key ) comment = `Standard, expl. primary t.key` )
( name = `it_std_w_std_pr_key_w_sec_key` itab_ref = REF #( it_std_w_std_pr_key_w_sec_key ) lines = lines( it_std_w_std_pr_key_w_sec_key ) comment = `Standard, standard primary, w. secondary t.key` )
( name = `it_std_w_emp_pr_key_w_sec_key` itab_ref = REF #( it_std_w_emp_pr_key_w_sec_key ) lines = lines( it_std_w_emp_pr_key_w_sec_key ) comment = `Standard, empty primary, w. secondary t.key` )
( name = `it_sorted` itab_ref = REF #( it_sorted ) lines = lines( it_sorted ) comment = `Sorted, primary t.key only` )
( name = `it_sorted_w_sec_key` itab_ref = REF #( it_sorted_w_sec_key ) lines = lines( it_sorted_w_sec_key ) comment = `Sorted, w. secondary t.key` )
( name = `it_hashed` itab_ref = REF #( it_hashed ) lines = lines( it_hashed ) comment = `Hashed, primary t.key only` )
( name = `it_hashed_w_sec_key` itab_ref = REF #( it_hashed_w_sec_key ) lines = lines( it_hashed_w_sec_key ) comment = `Hashed, w. secondary t.key` )
).
ENDMETHOD.
METHOD read_itabs.
FIELD-SYMBOLS <itab_idx> TYPE INDEX TABLE.
FIELD-SYMBOLS <itab_ha> TYPE HASHED TABLE.
FIELD-SYMBOLS <any_tab> TYPE ANY TABLE.
DATA ts1 TYPE utclong.
DATA ts2 TYPE utclong.
DATA seconds TYPE decfloat34.
DATA result TYPE info_tab_type.
IF info_tab IS INITIAL.
RETURN.
ELSE.
ASSIGN info_tab[ 1 ]-itab_ref->* TO FIELD-SYMBOL(<tab>).
"The number of read lines should not exceed the number of read accesses.
IF number_of_reads > lines( <tab> ).
RETURN.
ENDIF.
ENDIF.
"Repeated runtime check for a more meaningful result
DO number_of_repetitions TIMES.
LOOP AT info_tab REFERENCE INTO DATA(dref).
dref->id = sy-tabix.
"Assigning number of read accesses
dref->accesses = number_of_reads.
"Getting type information (RTTI)
tdo = CAST cl_abap_tabledescr( cl_abap_typedescr=>describe_by_data( dref->itab_ref->* ) ).
"Adding table kind information
"S: standard
"H: hashed
"O: sorted
dref->table_kind = tdo->table_kind.
"Adding table key information
dref->keys = tdo->get_keys( ).
"Assigning referenced internal tables to field symbols to work
"with all kinds of tables in the read accesses
IF dref->table_kind = 'H'.
ASSIGN dref->itab_ref->* TO <itab_ha>.
ASSIGN dref->itab_ref->* TO <any_tab>.
ELSE.
ASSIGN dref->itab_ref->* TO <itab_idx>.
ASSIGN dref->itab_ref->* TO <any_tab>.
ENDIF.
"-------------------- Read access by primary table index --------------------
"Only index tables have a primary table index
IF <itab_idx> IS ASSIGNED.
dref->operation = |Read access by primary table index|.
ts1 = utclong_current( ).
DO number_of_reads TIMES.
"Note: The READ TABLE statements in the statements use the
"TRANSPORTING NO FIELDS addition to only focus on the read
"access, not storing the content.
READ TABLE <itab_idx> INDEX sy-index TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
dref->runtime = seconds.
APPEND dref->* TO result.
ENDIF.
"-------------------- Read access by secondary table index --------------------
"Checking whether there is a secondary table key, which is indicated by an initial
"value of the 'is_primary' field.
DATA(has_sec_table_index) = xsdbool( line_exists( dref->keys[ is_primary = '' ] ) ).
IF has_sec_table_index = abap_true.
dref->operation = |Read access by secondary table index|.
"Getting a secondary key name for the dynamic READ TABLE
"statement below
DATA(sec_key_name) = dref->keys[ is_primary = '' ]-name.
ts1 = utclong_current( ).
DO number_of_reads TIMES.
READ TABLE <any_tab> INDEX sy-index USING KEY (sec_key_name) TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
dref->runtime = seconds.
APPEND dref->* TO result.
ENDIF.
"-------------------- Read access by primary table key --------------------
"Checking whether the internal table has a primary table key
READ TABLE dref->keys WITH KEY is_primary = 'X' REFERENCE INTO DATA(line_ref).
IF sy-subrc = 0.
"The example is designed for this example's demo internal tables. These
"tables all have a specific line type. In this case, the primary table
"key should only consist of one component and have the name 'IDX'.
DATA(key_check_lines) = xsdbool( lines( line_ref->components ) = 1 ).
IF key_check_lines = abap_true.
DATA(key_check_idx) = xsdbool( line_ref->components[ 1 ]-name = 'IDX' ).
ENDIF.
DATA(is_pr_key_only_idx) = xsdbool( key_check_lines = abap_true AND key_check_idx = abap_true ).
ELSE.
is_pr_key_only_idx = abap_false.
ENDIF.
IF is_pr_key_only_idx = abap_true.
dref->operation = |Read access by primary table key|.
ts1 = utclong_current( ).
DO number_of_reads TIMES.
"Note: The primary table key is always accessible by the default name 'primary_key'.
READ TABLE <any_tab> WITH TABLE KEY primary_key COMPONENTS ('IDX') = sy-index TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
dref->runtime = seconds.
APPEND dref->* TO result.
ENDIF.
"-------------------- Read access by secondary table key --------------------
"Checking whether the internal table has a secondary table key
READ TABLE dref->keys WITH KEY is_primary = '' REFERENCE INTO DATA(line_ref_sec).
IF sy-subrc = 0.
"The example is designed for this example's demo internal tables. These
"tables all have a specific line type. All tables with a secondary
"table key have one component with the name 'NUM'.
DATA(name_sec_key) = line_ref_sec->name.
DATA(sec_key_check_lines) = xsdbool( lines( line_ref_sec->components ) = 1 ).
IF sec_key_check_lines = abap_true.
DATA(sec_key_check_num) = xsdbool( line_ref_sec->components[ 1 ]-name = 'NUM' ).
ENDIF.
DATA(is_sec_key_only_num) = xsdbool( sec_key_check_lines = abap_true AND sec_key_check_num = abap_true ).
ELSE.
is_sec_key_only_num = abap_false.
ENDIF.
IF is_sec_key_only_num = abap_true.
dref->operation = |Read access by secondary table key|.
ts1 = utclong_current( ).
DO number_of_reads TIMES.
READ TABLE <any_tab> WITH TABLE KEY (name_sec_key) COMPONENTS ('NUM') = sy-index TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
dref->runtime = seconds.
APPEND dref->* TO result.
ENDIF.
"-------------------- Read access by free key --------------------
"The free key can contain components other than the key components.
"In this example, the line type of the internal tables consists of three
"components. One of the components constitutes the primary table key.
"The free key used here consists of the other two components.
"Checking whether the internal table includes the two components
"that constitute the free key.
IF check_example_components( ).
dref->operation = |Read access by free key|.
ts1 = utclong_current( ).
DO number_of_reads TIMES.
READ TABLE <any_tab> WITH KEY ('TXT') = `INDEX` && sy-index ('NUM') = sy-index TRANSPORTING NO FIELDS.
ENDDO.
ts2 = utclong_current( ).
cl_abap_utclong=>diff( EXPORTING high = ts2
low = ts1
IMPORTING seconds = seconds ).
dref->runtime = seconds.
APPEND dref->* TO result.
ENDIF.
"---------- Read access to standard tables by free key, using BINARY SEARCH ----------
"The example is intended to compare the runtime results of standard tables
"declared with secondary table keys vs the access using BINARY SEARCH that
"enables optimized access. Note that it is recommended to use secondary table
"keys for an optimized access.
IF dref->table_kind = 'S' AND <itab_idx> IS ASSIGNED.
IF check_example_components( ).
dref->operation = |Read access by free key and BINARY SEARCH|.
ts1 = utclong_current( ).
"A prior sorting is required (which is actually not needed in this example because
"the internal tables are populated in a sorted order).
SORT <itab_idx> BY ('TXT') ('NUM') ASCENDING.
DO number_of_reads TIMES.
READ TABLE <itab_idx> WITH KEY ('TXT') = `INDEX` && sy-index ('NUM') = sy-index BINARY SEARCH TRANSPORTING NO FIELDS.
ENDDO.
cl_abap_utclong=>diff( EXPORTING high = utclong_current( )
low = ts1
IMPORTING seconds = seconds ).
dref->runtime = seconds.
APPEND dref->* TO result.
ENDIF.
ENDIF.
UNASSIGN: <itab_idx>, <itab_ha>, <any_tab>.
ENDLOOP.
ENDDO.
"Retrieving the fastest run from the result table and adding this run to
"the internal table that is returned
LOOP AT result INTO DATA(wa) GROUP BY ( key1 = wa-id key2 = wa-operation ) ASCENDING.
LOOP AT GROUP wa INTO DATA(member) GROUP BY member-runtime ASCENDING.
"Clearing internal table content for output purposes
CLEAR member-itab_ref.
APPEND member TO read_results.
EXIT.
ENDLOOP.
ENDLOOP.
SORT read_results BY runtime ASCENDING.
ENDMETHOD.
METHOD check_example_components.
"Checking whether tables contain the components 'TXT' and 'NUM'
DATA(tab_line_type) = tdo->get_table_line_type( ).
DATA(struc_comps) = CAST cl_abap_structdescr( tab_line_type )->components.
are_included = xsdbool( line_exists( struc_comps[ name = 'TXT' ] ) AND line_exists( struc_comps[ name = 'NUM' ] ) ).
ENDMETHOD.
ENDCLASS.
- Formal parameters of methods or field symbols can be specified with generic types.
- At runtime, the actual data type is copied from the assigned actual parameter or memory area, i.e. they receive the complete data type only when an actual parameter is passed or a memory area is assigned.
- Among them, there are generic table types. For more information, refer to the Data Types and Objects cheat sheet and the documentation.
- The following example mainly demonstrates formal parameters of methods that are typed with generic table types. The method calls and the tables passed are only possible if the generic types fit. For example, you cannot pass a hashed table to a method whose importing parameter is typed with the generic type
INDEX TABLE
. Invalid method calls and table passing are commented out.
CLASS zcl_some_class DEFINITION
PUBLIC
FINAL
CREATE PUBLIC .
PUBLIC SECTION.
INTERFACES if_oo_adt_classrun.
CLASS-METHODS process_any_table IMPORTING itab TYPE ANY TABLE.
CLASS-METHODS process_standard_tables1 IMPORTING itab TYPE STANDARD TABLE.
CLASS-METHODS process_standard_tables2 IMPORTING itab TYPE TABLE.
CLASS-METHODS process_sorted_tables IMPORTING itab TYPE SORTED TABLE.
CLASS-METHODS process_hashed_tables IMPORTING itab TYPE HASHED TABLE.
CLASS-METHODS process_index_tables IMPORTING itab TYPE INDEX TABLE.
ENDCLASS.
CLASS zcl_some_class IMPLEMENTATION.
METHOD if_oo_adt_classrun~main.
DATA standard_tab TYPE TABLE OF string WITH EMPTY KEY.
DATA sorted_tab TYPE SORTED TABLE OF string WITH NON-UNIQUE KEY table_line.
DATA hashed_tab TYPE HASHED TABLE OF string WITH UNIQUE KEY table_line.
"ANY TABLE
process_any_table( standard_tab ).
process_any_table( sorted_tab ).
process_any_table( hashed_tab ).
"(STANDARD) TABLE
process_standard_tables1( standard_tab ).
"process_standard_tables1( sorted_tab ).
"process_standard_tables1( hashed_tab ).
process_standard_tables2( standard_tab ).
"process_standard_tables2( sorted_tab ).
"process_standard_tables2( hashed_tab ).
"SORTED TABLE
"process_sorted_tables( standard_tab ).
process_sorted_tables( sorted_tab ).
"process_sorted_tables( hashed_tab ).
"HASHED TABLE
"process_hashed_tables( standard_tab ).
"process_hashed_tables( sorted_tab ).
process_hashed_tables( hashed_tab ).
"INDEX TABLE
process_index_tables( standard_tab ).
process_index_tables( sorted_tab ).
"process_index_tables( hashed_tab ).
"Note: Field symbols can also be typed with generic types.
FIELD-SYMBOLS <fs_std_table> TYPE table.
ASSIGN standard_tab TO <fs_std_table>.
"ASSIGN sorted_tab TO <fs_std_table>.
"ASSIGN hashed_tab TO <fs_std_table>.
FIELD-SYMBOLS <fs_index_table> TYPE INDEX TABLE.
ASSIGN standard_tab TO <fs_index_table>.
ASSIGN sorted_tab TO <fs_index_table>.
"ASSIGN hashed_tab TO <fs_index_table>.
ENDMETHOD.
METHOD process_any_table.
ENDMETHOD.
METHOD process_hashed_tables.
ENDMETHOD.
METHOD process_index_tables.
ENDMETHOD.
METHOD process_sorted_tables.
ENDMETHOD.
METHOD process_standard_tables1.
ENDMETHOD.
METHOD process_standard_tables2.
ENDMETHOD.
ENDCLASS.
You can use FIND ... IN TABLE
statements to search for substrings in internal tables (standard tables without secondary table keys; with character-like line type) line by line.
DATA(str_table) = VALUE string_table( ( `aZbzZ` ) ( `cdZze` ) ( `Zzzf` ) ( `ghz` ) ).
"Finding all occurrences in a table
"Note: res_tab is of type match_result_tab
"You can also restrict the search range in an internal table; see an example in REPLACE ... IN TABLE
FIND ALL OCCURRENCES OF `Z`
IN TABLE str_table
RESULTS DATA(res_tab)
RESPECTING CASE.
"4 entries in table res_tab (tables in SUBMATCHES are initial since no regular expression is used)
"1. line: 1, offset: 1, length: 1, submatches: (initial)
"2. line: 1, offset: 4, length: 1, ...
"3. line: 2, offset: 2, length: 1, ...
"4. line: 3, offset: 0, length: 1, ...
"Finding the first occurrence in a table
"Note: res_struc, which is declared inline here, is of type match_result
FIND FIRST OCCURRENCE OF `Z`
IN TABLE str_table
RESULTS DATA(res_struc)
RESPECTING CASE.
"Entries in structure res_struc
"line: 1, offset: 1, length: 1, submatches: (initial)
"Alternative to the statement above (storing the information in individual data objects)
FIND FIRST OCCURRENCE OF `Z`
IN TABLE str_table
MATCH LINE DATA(line) "1
MATCH OFFSET DATA(off) "1
MATCH LENGTH DATA(len) "1
RESPECTING CASE.
Replacements in internal tables with REPLACE ... IN TABLE
:
DATA(str_table_original) = VALUE string_table( ( `aZbzZ` ) ( `cdZze` ) ( `Zzzf` ) ( `ghz` ) ).
DATA(str_table) = str_table_original.
"Replacing all occurrences in a table
"RESULTS addition: Storing information in an internal table of type repl_result_tab
REPLACE ALL OCCURRENCES OF `Z`
IN TABLE str_table
WITH `#`
RESULTS DATA(res_table)
RESPECTING CASE.
"str_table: a#bz# / cd#ze / #zzf / ghz
"res_table:
"LINE OFFSET LENGTH
"1 1 1
"1 4 1
"2 2 1
"3 0 1
str_table = str_table_original.
"Replacing the first occurrence in a table
"RESULTS addition: Storing information in a structure of type repl_result
REPLACE FIRST OCCURRENCE OF `Z`
IN TABLE str_table
WITH `#`
RESULTS DATA(res_structure)
RESPECTING CASE.
"str_table: a#bzZ / cdZze / Zzzf / ghz
"res_structure:
"LINE OFFSET LENGTH
"1 1 1
str_table = str_table_original.
"Restricting the search range in an internal table
REPLACE ALL OCCURRENCES OF `Z`
IN TABLE str_table
FROM 1 TO 2
WITH `#`
RESPECTING CASE.
"str_table: a#bz# / cd#ze / Zzzf / ghz
str_table = str_table_original.
"Offsets can be optionally specified (also only the offset of start or end line possible)
REPLACE ALL OCCURRENCES OF `Z`
IN TABLE str_table
FROM 1 OFFSET 3 TO 2 OFFSET 2
WITH `#`
RESPECTING CASE.
"str_table: aZbz# / cdZze / Zzzf / ghz
- Internal tables that have the predefined columns
SIGN
,OPTION
,LOW
, andHIGH
SIGN
: type c with length 1; indicates whether each line is included (I
) or excluded (E
) from the result set.OPTION
: type c with length 2; specifies the selection option for the condition. It uses comparison operators such asEQ
,NE
,GE
,GT
,LE
,LT
,CP
,NP
,BT
,NB
. Special rules apply for certain operators such as when usingCP
orNP
, theLOW
andHIGH
columns must be character values.LOW
: The lower comparison value.HIGH
: The higher comparison value.
- Declared with the
TYPE RANGE OF
addition inDATA
andTYPES
statements - Used to store range conditions that can be evaluated in expressions using the
IN
operator (each row in the table represents a separate comparison)
"Populating an integer table with values from 1 to 20
TYPES int_tab_type TYPE TABLE OF i WITH EMPTY KEY.
DATA(inttab) = VALUE int_tab_type( FOR x = 1 WHILE x <= 20 ( x ) ).
"Declaring a ranges table
DATA rangestab TYPE RANGE OF i.
"Populating a ranges table using VALUE
rangestab = VALUE #( sign = 'I'
option = 'BT' ( low = 1 high = 3 )
( low = 6 high = 8 )
( low = 12 high = 15 )
option = 'GE' ( low = 18 ) ).
"Using a SELECT statement and the IN addition to retrieve internal table
"content based on the ranges table specifications
SELECT * FROM @inttab AS tab
WHERE table_line IN @rangestab
INTO TABLE @DATA(result).
"result: 1, 2, 3, 6, 7, 8, 12, 13, 14, 15, 18, 19, 20
Using the methods of the CL_ABAP_DIFF
class, you can compare the content of two compatible index tables.
Find ...
- more information in the class documentation and in the ABAP Keyword Documentation.
- a code snippet in the Misc ABAP Classes cheat sheet.
In the context of ABAP RAP, the operands of ABAP EML statements and parameters of RAP handler methods and RAP saver methods are mainly special messenger tables for passing data and receiving results or messages: BDEF derived types - special ABAP types (internal tables and structures) that are tailor-made for RAP purposes.
"Example declarations
"For an EML create operation
DATA create_tab TYPE TABLE FOR CREATE entity.
"For an update operation
DATA update_tab TYPE TABLE FOR UPDATE entity.
"Type declaration using a BDEF derived type
TYPES der_typ TYPE TABLE FOR DELETE entity.
Find more information in the ABAP for RAP: Entity Manipulation Language (ABAP EML) ABAP cheat sheet.
Topic Internal Tables in the ABAP Keyword Documentation.
💡 Note
- The executable example covers the following topics, among others: Creating, populating, reading from, sorting, modifying internal tables
- The steps to import and run the code are outlined here.
- Disclaimer